Merge branch 'jc/finalize-temp-file'
[git/git.git] / sha1_file.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "string-list.h"
11 #include "lockfile.h"
12 #include "delta.h"
13 #include "pack.h"
14 #include "blob.h"
15 #include "commit.h"
16 #include "run-command.h"
17 #include "tag.h"
18 #include "tree.h"
19 #include "tree-walk.h"
20 #include "refs.h"
21 #include "pack-revindex.h"
22 #include "sha1-lookup.h"
23 #include "bulk-checkin.h"
24 #include "streaming.h"
25 #include "dir.h"
26
27 #ifndef O_NOATIME
28 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29 #define O_NOATIME 01000000
30 #else
31 #define O_NOATIME 0
32 #endif
33 #endif
34
35 #define SZ_FMT PRIuMAX
36 static inline uintmax_t sz_fmt(size_t s) { return s; }
37
38 const unsigned char null_sha1[20];
39
40 /*
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
44 * application).
45 */
46 static struct cached_object {
47 unsigned char sha1[20];
48 enum object_type type;
49 void *buf;
50 unsigned long size;
51 } *cached_objects;
52 static int cached_object_nr, cached_object_alloc;
53
54 static struct cached_object empty_tree = {
55 EMPTY_TREE_SHA1_BIN_LITERAL,
56 OBJ_TREE,
57 "",
58 0
59 };
60
61 /*
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
66 */
67 static struct packed_git *last_found_pack;
68
69 static struct cached_object *find_cached_object(const unsigned char *sha1)
70 {
71 int i;
72 struct cached_object *co = cached_objects;
73
74 for (i = 0; i < cached_object_nr; i++, co++) {
75 if (!hashcmp(co->sha1, sha1))
76 return co;
77 }
78 if (!hashcmp(sha1, empty_tree.sha1))
79 return &empty_tree;
80 return NULL;
81 }
82
83 int mkdir_in_gitdir(const char *path)
84 {
85 if (mkdir(path, 0777)) {
86 int saved_errno = errno;
87 struct stat st;
88 struct strbuf sb = STRBUF_INIT;
89
90 if (errno != EEXIST)
91 return -1;
92 /*
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
98 */
99 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
100 strbuf_readlink(&sb, path, st.st_size) ||
101 !is_absolute_path(sb.buf) ||
102 mkdir(sb.buf, 0777)) {
103 strbuf_release(&sb);
104 errno = saved_errno;
105 return -1;
106 }
107 strbuf_release(&sb);
108 }
109 return adjust_shared_perm(path);
110 }
111
112 enum scld_error safe_create_leading_directories(char *path)
113 {
114 char *next_component = path + offset_1st_component(path);
115 enum scld_error ret = SCLD_OK;
116
117 while (ret == SCLD_OK && next_component) {
118 struct stat st;
119 char *slash = next_component, slash_character;
120
121 while (*slash && !is_dir_sep(*slash))
122 slash++;
123
124 if (!*slash)
125 break;
126
127 next_component = slash + 1;
128 while (is_dir_sep(*next_component))
129 next_component++;
130 if (!*next_component)
131 break;
132
133 slash_character = *slash;
134 *slash = '\0';
135 if (!stat(path, &st)) {
136 /* path exists */
137 if (!S_ISDIR(st.st_mode))
138 ret = SCLD_EXISTS;
139 } else if (mkdir(path, 0777)) {
140 if (errno == EEXIST &&
141 !stat(path, &st) && S_ISDIR(st.st_mode))
142 ; /* somebody created it since we checked */
143 else if (errno == ENOENT)
144 /*
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
151 * trying again:
152 */
153 ret = SCLD_VANISHED;
154 else
155 ret = SCLD_FAILED;
156 } else if (adjust_shared_perm(path)) {
157 ret = SCLD_PERMS;
158 }
159 *slash = slash_character;
160 }
161 return ret;
162 }
163
164 enum scld_error safe_create_leading_directories_const(const char *path)
165 {
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf = xstrdup(path);
168 enum scld_error result = safe_create_leading_directories(buf);
169 free(buf);
170 return result;
171 }
172
173 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
174 {
175 int i;
176 for (i = 0; i < 20; i++) {
177 static char hex[] = "0123456789abcdef";
178 unsigned int val = sha1[i];
179 char *pos = pathbuf + i*2 + (i > 0);
180 *pos++ = hex[val >> 4];
181 *pos = hex[val & 0xf];
182 }
183 }
184
185 const char *sha1_file_name(const unsigned char *sha1)
186 {
187 static char buf[PATH_MAX];
188 const char *objdir;
189 int len;
190
191 objdir = get_object_directory();
192 len = strlen(objdir);
193
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len + 43 > PATH_MAX)
196 die("insanely long object directory %s", objdir);
197 memcpy(buf, objdir, len);
198 buf[len] = '/';
199 buf[len+3] = '/';
200 buf[len+42] = '\0';
201 fill_sha1_path(buf + len + 1, sha1);
202 return buf;
203 }
204
205 /*
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
209 */
210 static char *sha1_get_pack_name(const unsigned char *sha1,
211 char **name, char **base, const char *which)
212 {
213 static const char hex[] = "0123456789abcdef";
214 char *buf;
215 int i;
216
217 if (!*base) {
218 const char *sha1_file_directory = get_object_directory();
219 int len = strlen(sha1_file_directory);
220 *base = xmalloc(len + 60);
221 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory, which);
223 *name = *base + len + 11;
224 }
225
226 buf = *name;
227
228 for (i = 0; i < 20; i++) {
229 unsigned int val = *sha1++;
230 *buf++ = hex[val >> 4];
231 *buf++ = hex[val & 0xf];
232 }
233
234 return *base;
235 }
236
237 char *sha1_pack_name(const unsigned char *sha1)
238 {
239 static char *name, *base;
240
241 return sha1_get_pack_name(sha1, &name, &base, "pack");
242 }
243
244 char *sha1_pack_index_name(const unsigned char *sha1)
245 {
246 static char *name, *base;
247
248 return sha1_get_pack_name(sha1, &name, &base, "idx");
249 }
250
251 struct alternate_object_database *alt_odb_list;
252 static struct alternate_object_database **alt_odb_tail;
253
254 /*
255 * Prepare alternate object database registry.
256 *
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
267 * terminating NUL.
268 */
269 static int link_alt_odb_entry(const char *entry, const char *relative_base,
270 int depth, const char *normalized_objdir)
271 {
272 struct alternate_object_database *ent;
273 struct alternate_object_database *alt;
274 int pfxlen, entlen;
275 struct strbuf pathbuf = STRBUF_INIT;
276
277 if (!is_absolute_path(entry) && relative_base) {
278 strbuf_addstr(&pathbuf, real_path(relative_base));
279 strbuf_addch(&pathbuf, '/');
280 }
281 strbuf_addstr(&pathbuf, entry);
282
283 normalize_path_copy(pathbuf.buf, pathbuf.buf);
284
285 pfxlen = strlen(pathbuf.buf);
286
287 /*
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
290 */
291 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
292 pfxlen -= 1;
293
294 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent = xmalloc(sizeof(*ent) + entlen);
296 memcpy(ent->base, pathbuf.buf, pfxlen);
297 strbuf_release(&pathbuf);
298
299 ent->name = ent->base + pfxlen + 1;
300 ent->base[pfxlen + 3] = '/';
301 ent->base[pfxlen] = ent->base[entlen-1] = 0;
302
303 /* Detect cases where alternate disappeared */
304 if (!is_directory(ent->base)) {
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
307 ent->base);
308 free(ent);
309 return -1;
310 }
311
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
314 */
315 for (alt = alt_odb_list; alt; alt = alt->next) {
316 if (pfxlen == alt->name - alt->base - 1 &&
317 !memcmp(ent->base, alt->base, pfxlen)) {
318 free(ent);
319 return -1;
320 }
321 }
322 if (!strcmp_icase(ent->base, normalized_objdir)) {
323 free(ent);
324 return -1;
325 }
326
327 /* add the alternate entry */
328 *alt_odb_tail = ent;
329 alt_odb_tail = &(ent->next);
330 ent->next = NULL;
331
332 /* recursively add alternates */
333 read_info_alternates(ent->base, depth + 1);
334
335 ent->base[pfxlen] = '/';
336
337 return 0;
338 }
339
340 static void link_alt_odb_entries(const char *alt, int len, int sep,
341 const char *relative_base, int depth)
342 {
343 struct string_list entries = STRING_LIST_INIT_NODUP;
344 char *alt_copy;
345 int i;
346 struct strbuf objdirbuf = STRBUF_INIT;
347
348 if (depth > 5) {
349 error("%s: ignoring alternate object stores, nesting too deep.",
350 relative_base);
351 return;
352 }
353
354 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
355 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
356
357 alt_copy = xmemdupz(alt, len);
358 string_list_split_in_place(&entries, alt_copy, sep, -1);
359 for (i = 0; i < entries.nr; i++) {
360 const char *entry = entries.items[i].string;
361 if (entry[0] == '\0' || entry[0] == '#')
362 continue;
363 if (!is_absolute_path(entry) && depth) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base, entry);
366 } else {
367 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
368 }
369 }
370 string_list_clear(&entries, 0);
371 free(alt_copy);
372 strbuf_release(&objdirbuf);
373 }
374
375 void read_info_alternates(const char * relative_base, int depth)
376 {
377 char *map;
378 size_t mapsz;
379 struct stat st;
380 const char alt_file_name[] = "info/alternates";
381 /* Given that relative_base is no longer than PATH_MAX,
382 ensure that "path" has enough space to append "/", the
383 file name, "info/alternates", and a trailing NUL. */
384 char path[PATH_MAX + 1 + sizeof alt_file_name];
385 int fd;
386
387 sprintf(path, "%s/%s", relative_base, alt_file_name);
388 fd = git_open_noatime(path);
389 if (fd < 0)
390 return;
391 if (fstat(fd, &st) || (st.st_size == 0)) {
392 close(fd);
393 return;
394 }
395 mapsz = xsize_t(st.st_size);
396 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
397 close(fd);
398
399 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
400
401 munmap(map, mapsz);
402 }
403
404 void add_to_alternates_file(const char *reference)
405 {
406 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
407 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
408 const char *alt = mkpath("%s\n", reference);
409 write_or_die(fd, alt, strlen(alt));
410 if (commit_lock_file(lock))
411 die("could not close alternates file");
412 if (alt_odb_tail)
413 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
414 }
415
416 int foreach_alt_odb(alt_odb_fn fn, void *cb)
417 {
418 struct alternate_object_database *ent;
419 int r = 0;
420
421 prepare_alt_odb();
422 for (ent = alt_odb_list; ent; ent = ent->next) {
423 r = fn(ent, cb);
424 if (r)
425 break;
426 }
427 return r;
428 }
429
430 void prepare_alt_odb(void)
431 {
432 const char *alt;
433
434 if (alt_odb_tail)
435 return;
436
437 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
438 if (!alt) alt = "";
439
440 alt_odb_tail = &alt_odb_list;
441 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
442
443 read_info_alternates(get_object_directory(), 0);
444 }
445
446 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
447 static int freshen_file(const char *fn)
448 {
449 struct utimbuf t;
450 t.actime = t.modtime = time(NULL);
451 return !utime(fn, &t);
452 }
453
454 /*
455 * All of the check_and_freshen functions return 1 if the file exists and was
456 * freshened (if freshening was requested), 0 otherwise. If they return
457 * 0, you should not assume that it is safe to skip a write of the object (it
458 * either does not exist on disk, or has a stale mtime and may be subject to
459 * pruning).
460 */
461 static int check_and_freshen_file(const char *fn, int freshen)
462 {
463 if (access(fn, F_OK))
464 return 0;
465 if (freshen && !freshen_file(fn))
466 return 0;
467 return 1;
468 }
469
470 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
471 {
472 return check_and_freshen_file(sha1_file_name(sha1), freshen);
473 }
474
475 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
476 {
477 struct alternate_object_database *alt;
478 prepare_alt_odb();
479 for (alt = alt_odb_list; alt; alt = alt->next) {
480 fill_sha1_path(alt->name, sha1);
481 if (check_and_freshen_file(alt->base, freshen))
482 return 1;
483 }
484 return 0;
485 }
486
487 static int check_and_freshen(const unsigned char *sha1, int freshen)
488 {
489 return check_and_freshen_local(sha1, freshen) ||
490 check_and_freshen_nonlocal(sha1, freshen);
491 }
492
493 int has_loose_object_nonlocal(const unsigned char *sha1)
494 {
495 return check_and_freshen_nonlocal(sha1, 0);
496 }
497
498 static int has_loose_object(const unsigned char *sha1)
499 {
500 return check_and_freshen(sha1, 0);
501 }
502
503 static unsigned int pack_used_ctr;
504 static unsigned int pack_mmap_calls;
505 static unsigned int peak_pack_open_windows;
506 static unsigned int pack_open_windows;
507 static unsigned int pack_open_fds;
508 static unsigned int pack_max_fds;
509 static size_t peak_pack_mapped;
510 static size_t pack_mapped;
511 struct packed_git *packed_git;
512
513 void pack_report(void)
514 {
515 fprintf(stderr,
516 "pack_report: getpagesize() = %10" SZ_FMT "\n"
517 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
518 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
519 sz_fmt(getpagesize()),
520 sz_fmt(packed_git_window_size),
521 sz_fmt(packed_git_limit));
522 fprintf(stderr,
523 "pack_report: pack_used_ctr = %10u\n"
524 "pack_report: pack_mmap_calls = %10u\n"
525 "pack_report: pack_open_windows = %10u / %10u\n"
526 "pack_report: pack_mapped = "
527 "%10" SZ_FMT " / %10" SZ_FMT "\n",
528 pack_used_ctr,
529 pack_mmap_calls,
530 pack_open_windows, peak_pack_open_windows,
531 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
532 }
533
534 /*
535 * Open and mmap the index file at path, perform a couple of
536 * consistency checks, then record its information to p. Return 0 on
537 * success.
538 */
539 static int check_packed_git_idx(const char *path, struct packed_git *p)
540 {
541 void *idx_map;
542 struct pack_idx_header *hdr;
543 size_t idx_size;
544 uint32_t version, nr, i, *index;
545 int fd = git_open_noatime(path);
546 struct stat st;
547
548 if (fd < 0)
549 return -1;
550 if (fstat(fd, &st)) {
551 close(fd);
552 return -1;
553 }
554 idx_size = xsize_t(st.st_size);
555 if (idx_size < 4 * 256 + 20 + 20) {
556 close(fd);
557 return error("index file %s is too small", path);
558 }
559 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
560 close(fd);
561
562 hdr = idx_map;
563 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
564 version = ntohl(hdr->idx_version);
565 if (version < 2 || version > 2) {
566 munmap(idx_map, idx_size);
567 return error("index file %s is version %"PRIu32
568 " and is not supported by this binary"
569 " (try upgrading GIT to a newer version)",
570 path, version);
571 }
572 } else
573 version = 1;
574
575 nr = 0;
576 index = idx_map;
577 if (version > 1)
578 index += 2; /* skip index header */
579 for (i = 0; i < 256; i++) {
580 uint32_t n = ntohl(index[i]);
581 if (n < nr) {
582 munmap(idx_map, idx_size);
583 return error("non-monotonic index %s", path);
584 }
585 nr = n;
586 }
587
588 if (version == 1) {
589 /*
590 * Total size:
591 * - 256 index entries 4 bytes each
592 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
593 * - 20-byte SHA1 of the packfile
594 * - 20-byte SHA1 file checksum
595 */
596 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
597 munmap(idx_map, idx_size);
598 return error("wrong index v1 file size in %s", path);
599 }
600 } else if (version == 2) {
601 /*
602 * Minimum size:
603 * - 8 bytes of header
604 * - 256 index entries 4 bytes each
605 * - 20-byte sha1 entry * nr
606 * - 4-byte crc entry * nr
607 * - 4-byte offset entry * nr
608 * - 20-byte SHA1 of the packfile
609 * - 20-byte SHA1 file checksum
610 * And after the 4-byte offset table might be a
611 * variable sized table containing 8-byte entries
612 * for offsets larger than 2^31.
613 */
614 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
615 unsigned long max_size = min_size;
616 if (nr)
617 max_size += (nr - 1)*8;
618 if (idx_size < min_size || idx_size > max_size) {
619 munmap(idx_map, idx_size);
620 return error("wrong index v2 file size in %s", path);
621 }
622 if (idx_size != min_size &&
623 /*
624 * make sure we can deal with large pack offsets.
625 * 31-bit signed offset won't be enough, neither
626 * 32-bit unsigned one will be.
627 */
628 (sizeof(off_t) <= 4)) {
629 munmap(idx_map, idx_size);
630 return error("pack too large for current definition of off_t in %s", path);
631 }
632 }
633
634 p->index_version = version;
635 p->index_data = idx_map;
636 p->index_size = idx_size;
637 p->num_objects = nr;
638 return 0;
639 }
640
641 int open_pack_index(struct packed_git *p)
642 {
643 char *idx_name;
644 int ret;
645
646 if (p->index_data)
647 return 0;
648
649 idx_name = xstrdup(p->pack_name);
650 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
651 ret = check_packed_git_idx(idx_name, p);
652 free(idx_name);
653 return ret;
654 }
655
656 static void scan_windows(struct packed_git *p,
657 struct packed_git **lru_p,
658 struct pack_window **lru_w,
659 struct pack_window **lru_l)
660 {
661 struct pack_window *w, *w_l;
662
663 for (w_l = NULL, w = p->windows; w; w = w->next) {
664 if (!w->inuse_cnt) {
665 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
666 *lru_p = p;
667 *lru_w = w;
668 *lru_l = w_l;
669 }
670 }
671 w_l = w;
672 }
673 }
674
675 static int unuse_one_window(struct packed_git *current)
676 {
677 struct packed_git *p, *lru_p = NULL;
678 struct pack_window *lru_w = NULL, *lru_l = NULL;
679
680 if (current)
681 scan_windows(current, &lru_p, &lru_w, &lru_l);
682 for (p = packed_git; p; p = p->next)
683 scan_windows(p, &lru_p, &lru_w, &lru_l);
684 if (lru_p) {
685 munmap(lru_w->base, lru_w->len);
686 pack_mapped -= lru_w->len;
687 if (lru_l)
688 lru_l->next = lru_w->next;
689 else
690 lru_p->windows = lru_w->next;
691 free(lru_w);
692 pack_open_windows--;
693 return 1;
694 }
695 return 0;
696 }
697
698 void release_pack_memory(size_t need)
699 {
700 size_t cur = pack_mapped;
701 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
702 ; /* nothing */
703 }
704
705 static void mmap_limit_check(size_t length)
706 {
707 static size_t limit = 0;
708 if (!limit) {
709 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
710 if (!limit)
711 limit = SIZE_MAX;
712 }
713 if (length > limit)
714 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
715 (uintmax_t)length, (uintmax_t)limit);
716 }
717
718 void *xmmap_gently(void *start, size_t length,
719 int prot, int flags, int fd, off_t offset)
720 {
721 void *ret;
722
723 mmap_limit_check(length);
724 ret = mmap(start, length, prot, flags, fd, offset);
725 if (ret == MAP_FAILED) {
726 if (!length)
727 return NULL;
728 release_pack_memory(length);
729 ret = mmap(start, length, prot, flags, fd, offset);
730 }
731 return ret;
732 }
733
734 void *xmmap(void *start, size_t length,
735 int prot, int flags, int fd, off_t offset)
736 {
737 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
738 if (ret == MAP_FAILED)
739 die_errno("mmap failed");
740 return ret;
741 }
742
743 void close_pack_windows(struct packed_git *p)
744 {
745 while (p->windows) {
746 struct pack_window *w = p->windows;
747
748 if (w->inuse_cnt)
749 die("pack '%s' still has open windows to it",
750 p->pack_name);
751 munmap(w->base, w->len);
752 pack_mapped -= w->len;
753 pack_open_windows--;
754 p->windows = w->next;
755 free(w);
756 }
757 }
758
759 /*
760 * The LRU pack is the one with the oldest MRU window, preferring packs
761 * with no used windows, or the oldest mtime if it has no windows allocated.
762 */
763 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
764 {
765 struct pack_window *w, *this_mru_w;
766 int has_windows_inuse = 0;
767
768 /*
769 * Reject this pack if it has windows and the previously selected
770 * one does not. If this pack does not have windows, reject
771 * it if the pack file is newer than the previously selected one.
772 */
773 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
774 return;
775
776 for (w = this_mru_w = p->windows; w; w = w->next) {
777 /*
778 * Reject this pack if any of its windows are in use,
779 * but the previously selected pack did not have any
780 * inuse windows. Otherwise, record that this pack
781 * has windows in use.
782 */
783 if (w->inuse_cnt) {
784 if (*accept_windows_inuse)
785 has_windows_inuse = 1;
786 else
787 return;
788 }
789
790 if (w->last_used > this_mru_w->last_used)
791 this_mru_w = w;
792
793 /*
794 * Reject this pack if it has windows that have been
795 * used more recently than the previously selected pack.
796 * If the previously selected pack had windows inuse and
797 * we have not encountered a window in this pack that is
798 * inuse, skip this check since we prefer a pack with no
799 * inuse windows to one that has inuse windows.
800 */
801 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
802 this_mru_w->last_used > (*mru_w)->last_used)
803 return;
804 }
805
806 /*
807 * Select this pack.
808 */
809 *mru_w = this_mru_w;
810 *lru_p = p;
811 *accept_windows_inuse = has_windows_inuse;
812 }
813
814 static int close_one_pack(void)
815 {
816 struct packed_git *p, *lru_p = NULL;
817 struct pack_window *mru_w = NULL;
818 int accept_windows_inuse = 1;
819
820 for (p = packed_git; p; p = p->next) {
821 if (p->pack_fd == -1)
822 continue;
823 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
824 }
825
826 if (lru_p) {
827 close(lru_p->pack_fd);
828 pack_open_fds--;
829 lru_p->pack_fd = -1;
830 return 1;
831 }
832
833 return 0;
834 }
835
836 void unuse_pack(struct pack_window **w_cursor)
837 {
838 struct pack_window *w = *w_cursor;
839 if (w) {
840 w->inuse_cnt--;
841 *w_cursor = NULL;
842 }
843 }
844
845 void close_pack_index(struct packed_git *p)
846 {
847 if (p->index_data) {
848 munmap((void *)p->index_data, p->index_size);
849 p->index_data = NULL;
850 }
851 }
852
853 /*
854 * This is used by git-repack in case a newly created pack happens to
855 * contain the same set of objects as an existing one. In that case
856 * the resulting file might be different even if its name would be the
857 * same. It is best to close any reference to the old pack before it is
858 * replaced on disk. Of course no index pointers or windows for given pack
859 * must subsist at this point. If ever objects from this pack are requested
860 * again, the new version of the pack will be reinitialized through
861 * reprepare_packed_git().
862 */
863 void free_pack_by_name(const char *pack_name)
864 {
865 struct packed_git *p, **pp = &packed_git;
866
867 while (*pp) {
868 p = *pp;
869 if (strcmp(pack_name, p->pack_name) == 0) {
870 clear_delta_base_cache();
871 close_pack_windows(p);
872 if (p->pack_fd != -1) {
873 close(p->pack_fd);
874 pack_open_fds--;
875 }
876 close_pack_index(p);
877 free(p->bad_object_sha1);
878 *pp = p->next;
879 if (last_found_pack == p)
880 last_found_pack = NULL;
881 free(p);
882 return;
883 }
884 pp = &p->next;
885 }
886 }
887
888 static unsigned int get_max_fd_limit(void)
889 {
890 #ifdef RLIMIT_NOFILE
891 {
892 struct rlimit lim;
893
894 if (!getrlimit(RLIMIT_NOFILE, &lim))
895 return lim.rlim_cur;
896 }
897 #endif
898
899 #ifdef _SC_OPEN_MAX
900 {
901 long open_max = sysconf(_SC_OPEN_MAX);
902 if (0 < open_max)
903 return open_max;
904 /*
905 * Otherwise, we got -1 for one of the two
906 * reasons:
907 *
908 * (1) sysconf() did not understand _SC_OPEN_MAX
909 * and signaled an error with -1; or
910 * (2) sysconf() said there is no limit.
911 *
912 * We _could_ clear errno before calling sysconf() to
913 * tell these two cases apart and return a huge number
914 * in the latter case to let the caller cap it to a
915 * value that is not so selfish, but letting the
916 * fallback OPEN_MAX codepath take care of these cases
917 * is a lot simpler.
918 */
919 }
920 #endif
921
922 #ifdef OPEN_MAX
923 return OPEN_MAX;
924 #else
925 return 1; /* see the caller ;-) */
926 #endif
927 }
928
929 /*
930 * Do not call this directly as this leaks p->pack_fd on error return;
931 * call open_packed_git() instead.
932 */
933 static int open_packed_git_1(struct packed_git *p)
934 {
935 struct stat st;
936 struct pack_header hdr;
937 unsigned char sha1[20];
938 unsigned char *idx_sha1;
939 long fd_flag;
940
941 if (!p->index_data && open_pack_index(p))
942 return error("packfile %s index unavailable", p->pack_name);
943
944 if (!pack_max_fds) {
945 unsigned int max_fds = get_max_fd_limit();
946
947 /* Save 3 for stdin/stdout/stderr, 22 for work */
948 if (25 < max_fds)
949 pack_max_fds = max_fds - 25;
950 else
951 pack_max_fds = 1;
952 }
953
954 while (pack_max_fds <= pack_open_fds && close_one_pack())
955 ; /* nothing */
956
957 p->pack_fd = git_open_noatime(p->pack_name);
958 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
959 return -1;
960 pack_open_fds++;
961
962 /* If we created the struct before we had the pack we lack size. */
963 if (!p->pack_size) {
964 if (!S_ISREG(st.st_mode))
965 return error("packfile %s not a regular file", p->pack_name);
966 p->pack_size = st.st_size;
967 } else if (p->pack_size != st.st_size)
968 return error("packfile %s size changed", p->pack_name);
969
970 /* We leave these file descriptors open with sliding mmap;
971 * there is no point keeping them open across exec(), though.
972 */
973 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
974 if (fd_flag < 0)
975 return error("cannot determine file descriptor flags");
976 fd_flag |= FD_CLOEXEC;
977 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
978 return error("cannot set FD_CLOEXEC");
979
980 /* Verify we recognize this pack file format. */
981 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
982 return error("file %s is far too short to be a packfile", p->pack_name);
983 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
984 return error("file %s is not a GIT packfile", p->pack_name);
985 if (!pack_version_ok(hdr.hdr_version))
986 return error("packfile %s is version %"PRIu32" and not"
987 " supported (try upgrading GIT to a newer version)",
988 p->pack_name, ntohl(hdr.hdr_version));
989
990 /* Verify the pack matches its index. */
991 if (p->num_objects != ntohl(hdr.hdr_entries))
992 return error("packfile %s claims to have %"PRIu32" objects"
993 " while index indicates %"PRIu32" objects",
994 p->pack_name, ntohl(hdr.hdr_entries),
995 p->num_objects);
996 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
997 return error("end of packfile %s is unavailable", p->pack_name);
998 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
999 return error("packfile %s signature is unavailable", p->pack_name);
1000 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1001 if (hashcmp(sha1, idx_sha1))
1002 return error("packfile %s does not match index", p->pack_name);
1003 return 0;
1004 }
1005
1006 static int open_packed_git(struct packed_git *p)
1007 {
1008 if (!open_packed_git_1(p))
1009 return 0;
1010 if (p->pack_fd != -1) {
1011 close(p->pack_fd);
1012 pack_open_fds--;
1013 p->pack_fd = -1;
1014 }
1015 return -1;
1016 }
1017
1018 static int in_window(struct pack_window *win, off_t offset)
1019 {
1020 /* We must promise at least 20 bytes (one hash) after the
1021 * offset is available from this window, otherwise the offset
1022 * is not actually in this window and a different window (which
1023 * has that one hash excess) must be used. This is to support
1024 * the object header and delta base parsing routines below.
1025 */
1026 off_t win_off = win->offset;
1027 return win_off <= offset
1028 && (offset + 20) <= (win_off + win->len);
1029 }
1030
1031 unsigned char *use_pack(struct packed_git *p,
1032 struct pack_window **w_cursor,
1033 off_t offset,
1034 unsigned long *left)
1035 {
1036 struct pack_window *win = *w_cursor;
1037
1038 /* Since packfiles end in a hash of their content and it's
1039 * pointless to ask for an offset into the middle of that
1040 * hash, and the in_window function above wouldn't match
1041 * don't allow an offset too close to the end of the file.
1042 */
1043 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1044 die("packfile %s cannot be accessed", p->pack_name);
1045 if (offset > (p->pack_size - 20))
1046 die("offset beyond end of packfile (truncated pack?)");
1047
1048 if (!win || !in_window(win, offset)) {
1049 if (win)
1050 win->inuse_cnt--;
1051 for (win = p->windows; win; win = win->next) {
1052 if (in_window(win, offset))
1053 break;
1054 }
1055 if (!win) {
1056 size_t window_align = packed_git_window_size / 2;
1057 off_t len;
1058
1059 if (p->pack_fd == -1 && open_packed_git(p))
1060 die("packfile %s cannot be accessed", p->pack_name);
1061
1062 win = xcalloc(1, sizeof(*win));
1063 win->offset = (offset / window_align) * window_align;
1064 len = p->pack_size - win->offset;
1065 if (len > packed_git_window_size)
1066 len = packed_git_window_size;
1067 win->len = (size_t)len;
1068 pack_mapped += win->len;
1069 while (packed_git_limit < pack_mapped
1070 && unuse_one_window(p))
1071 ; /* nothing */
1072 win->base = xmmap(NULL, win->len,
1073 PROT_READ, MAP_PRIVATE,
1074 p->pack_fd, win->offset);
1075 if (win->base == MAP_FAILED)
1076 die("packfile %s cannot be mapped: %s",
1077 p->pack_name,
1078 strerror(errno));
1079 if (!win->offset && win->len == p->pack_size
1080 && !p->do_not_close) {
1081 close(p->pack_fd);
1082 pack_open_fds--;
1083 p->pack_fd = -1;
1084 }
1085 pack_mmap_calls++;
1086 pack_open_windows++;
1087 if (pack_mapped > peak_pack_mapped)
1088 peak_pack_mapped = pack_mapped;
1089 if (pack_open_windows > peak_pack_open_windows)
1090 peak_pack_open_windows = pack_open_windows;
1091 win->next = p->windows;
1092 p->windows = win;
1093 }
1094 }
1095 if (win != *w_cursor) {
1096 win->last_used = pack_used_ctr++;
1097 win->inuse_cnt++;
1098 *w_cursor = win;
1099 }
1100 offset -= win->offset;
1101 if (left)
1102 *left = win->len - xsize_t(offset);
1103 return win->base + offset;
1104 }
1105
1106 static struct packed_git *alloc_packed_git(int extra)
1107 {
1108 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1109 memset(p, 0, sizeof(*p));
1110 p->pack_fd = -1;
1111 return p;
1112 }
1113
1114 static void try_to_free_pack_memory(size_t size)
1115 {
1116 release_pack_memory(size);
1117 }
1118
1119 struct packed_git *add_packed_git(const char *path, int path_len, int local)
1120 {
1121 static int have_set_try_to_free_routine;
1122 struct stat st;
1123 struct packed_git *p = alloc_packed_git(path_len + 2);
1124
1125 if (!have_set_try_to_free_routine) {
1126 have_set_try_to_free_routine = 1;
1127 set_try_to_free_routine(try_to_free_pack_memory);
1128 }
1129
1130 /*
1131 * Make sure a corresponding .pack file exists and that
1132 * the index looks sane.
1133 */
1134 path_len -= strlen(".idx");
1135 if (path_len < 1) {
1136 free(p);
1137 return NULL;
1138 }
1139 memcpy(p->pack_name, path, path_len);
1140
1141 strcpy(p->pack_name + path_len, ".keep");
1142 if (!access(p->pack_name, F_OK))
1143 p->pack_keep = 1;
1144
1145 strcpy(p->pack_name + path_len, ".pack");
1146 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1147 free(p);
1148 return NULL;
1149 }
1150
1151 /* ok, it looks sane as far as we can check without
1152 * actually mapping the pack file.
1153 */
1154 p->pack_size = st.st_size;
1155 p->pack_local = local;
1156 p->mtime = st.st_mtime;
1157 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1158 hashclr(p->sha1);
1159 return p;
1160 }
1161
1162 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1163 {
1164 const char *path = sha1_pack_name(sha1);
1165 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1166
1167 strcpy(p->pack_name, path);
1168 hashcpy(p->sha1, sha1);
1169 if (check_packed_git_idx(idx_path, p)) {
1170 free(p);
1171 return NULL;
1172 }
1173
1174 return p;
1175 }
1176
1177 void install_packed_git(struct packed_git *pack)
1178 {
1179 if (pack->pack_fd != -1)
1180 pack_open_fds++;
1181
1182 pack->next = packed_git;
1183 packed_git = pack;
1184 }
1185
1186 void (*report_garbage)(const char *desc, const char *path);
1187
1188 static void report_helper(const struct string_list *list,
1189 int seen_bits, int first, int last)
1190 {
1191 const char *msg;
1192 switch (seen_bits) {
1193 case 0:
1194 msg = "no corresponding .idx or .pack";
1195 break;
1196 case 1:
1197 msg = "no corresponding .idx";
1198 break;
1199 case 2:
1200 msg = "no corresponding .pack";
1201 break;
1202 default:
1203 return;
1204 }
1205 for (; first < last; first++)
1206 report_garbage(msg, list->items[first].string);
1207 }
1208
1209 static void report_pack_garbage(struct string_list *list)
1210 {
1211 int i, baselen = -1, first = 0, seen_bits = 0;
1212
1213 if (!report_garbage)
1214 return;
1215
1216 string_list_sort(list);
1217
1218 for (i = 0; i < list->nr; i++) {
1219 const char *path = list->items[i].string;
1220 if (baselen != -1 &&
1221 strncmp(path, list->items[first].string, baselen)) {
1222 report_helper(list, seen_bits, first, i);
1223 baselen = -1;
1224 seen_bits = 0;
1225 }
1226 if (baselen == -1) {
1227 const char *dot = strrchr(path, '.');
1228 if (!dot) {
1229 report_garbage("garbage found", path);
1230 continue;
1231 }
1232 baselen = dot - path + 1;
1233 first = i;
1234 }
1235 if (!strcmp(path + baselen, "pack"))
1236 seen_bits |= 1;
1237 else if (!strcmp(path + baselen, "idx"))
1238 seen_bits |= 2;
1239 }
1240 report_helper(list, seen_bits, first, list->nr);
1241 }
1242
1243 static void prepare_packed_git_one(char *objdir, int local)
1244 {
1245 struct strbuf path = STRBUF_INIT;
1246 size_t dirnamelen;
1247 DIR *dir;
1248 struct dirent *de;
1249 struct string_list garbage = STRING_LIST_INIT_DUP;
1250
1251 strbuf_addstr(&path, objdir);
1252 strbuf_addstr(&path, "/pack");
1253 dir = opendir(path.buf);
1254 if (!dir) {
1255 if (errno != ENOENT)
1256 error("unable to open object pack directory: %s: %s",
1257 path.buf, strerror(errno));
1258 strbuf_release(&path);
1259 return;
1260 }
1261 strbuf_addch(&path, '/');
1262 dirnamelen = path.len;
1263 while ((de = readdir(dir)) != NULL) {
1264 struct packed_git *p;
1265 size_t base_len;
1266
1267 if (is_dot_or_dotdot(de->d_name))
1268 continue;
1269
1270 strbuf_setlen(&path, dirnamelen);
1271 strbuf_addstr(&path, de->d_name);
1272
1273 base_len = path.len;
1274 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1275 /* Don't reopen a pack we already have. */
1276 for (p = packed_git; p; p = p->next) {
1277 size_t len;
1278 if (strip_suffix(p->pack_name, ".pack", &len) &&
1279 len == base_len &&
1280 !memcmp(p->pack_name, path.buf, len))
1281 break;
1282 }
1283 if (p == NULL &&
1284 /*
1285 * See if it really is a valid .idx file with
1286 * corresponding .pack file that we can map.
1287 */
1288 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1289 install_packed_git(p);
1290 }
1291
1292 if (!report_garbage)
1293 continue;
1294
1295 if (ends_with(de->d_name, ".idx") ||
1296 ends_with(de->d_name, ".pack") ||
1297 ends_with(de->d_name, ".bitmap") ||
1298 ends_with(de->d_name, ".keep"))
1299 string_list_append(&garbage, path.buf);
1300 else
1301 report_garbage("garbage found", path.buf);
1302 }
1303 closedir(dir);
1304 report_pack_garbage(&garbage);
1305 string_list_clear(&garbage, 0);
1306 strbuf_release(&path);
1307 }
1308
1309 static int sort_pack(const void *a_, const void *b_)
1310 {
1311 struct packed_git *a = *((struct packed_git **)a_);
1312 struct packed_git *b = *((struct packed_git **)b_);
1313 int st;
1314
1315 /*
1316 * Local packs tend to contain objects specific to our
1317 * variant of the project than remote ones. In addition,
1318 * remote ones could be on a network mounted filesystem.
1319 * Favor local ones for these reasons.
1320 */
1321 st = a->pack_local - b->pack_local;
1322 if (st)
1323 return -st;
1324
1325 /*
1326 * Younger packs tend to contain more recent objects,
1327 * and more recent objects tend to get accessed more
1328 * often.
1329 */
1330 if (a->mtime < b->mtime)
1331 return 1;
1332 else if (a->mtime == b->mtime)
1333 return 0;
1334 return -1;
1335 }
1336
1337 static void rearrange_packed_git(void)
1338 {
1339 struct packed_git **ary, *p;
1340 int i, n;
1341
1342 for (n = 0, p = packed_git; p; p = p->next)
1343 n++;
1344 if (n < 2)
1345 return;
1346
1347 /* prepare an array of packed_git for easier sorting */
1348 ary = xcalloc(n, sizeof(struct packed_git *));
1349 for (n = 0, p = packed_git; p; p = p->next)
1350 ary[n++] = p;
1351
1352 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1353
1354 /* link them back again */
1355 for (i = 0; i < n - 1; i++)
1356 ary[i]->next = ary[i + 1];
1357 ary[n - 1]->next = NULL;
1358 packed_git = ary[0];
1359
1360 free(ary);
1361 }
1362
1363 static int prepare_packed_git_run_once = 0;
1364 void prepare_packed_git(void)
1365 {
1366 struct alternate_object_database *alt;
1367
1368 if (prepare_packed_git_run_once)
1369 return;
1370 prepare_packed_git_one(get_object_directory(), 1);
1371 prepare_alt_odb();
1372 for (alt = alt_odb_list; alt; alt = alt->next) {
1373 alt->name[-1] = 0;
1374 prepare_packed_git_one(alt->base, 0);
1375 alt->name[-1] = '/';
1376 }
1377 rearrange_packed_git();
1378 prepare_packed_git_run_once = 1;
1379 }
1380
1381 void reprepare_packed_git(void)
1382 {
1383 prepare_packed_git_run_once = 0;
1384 prepare_packed_git();
1385 }
1386
1387 static void mark_bad_packed_object(struct packed_git *p,
1388 const unsigned char *sha1)
1389 {
1390 unsigned i;
1391 for (i = 0; i < p->num_bad_objects; i++)
1392 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1393 return;
1394 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1395 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1396 p->num_bad_objects++;
1397 }
1398
1399 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1400 {
1401 struct packed_git *p;
1402 unsigned i;
1403
1404 for (p = packed_git; p; p = p->next)
1405 for (i = 0; i < p->num_bad_objects; i++)
1406 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1407 return p;
1408 return NULL;
1409 }
1410
1411 /*
1412 * With an in-core object data in "map", rehash it to make sure the
1413 * object name actually matches "sha1" to detect object corruption.
1414 * With "map" == NULL, try reading the object named with "sha1" using
1415 * the streaming interface and rehash it to do the same.
1416 */
1417 int check_sha1_signature(const unsigned char *sha1, void *map,
1418 unsigned long size, const char *type)
1419 {
1420 unsigned char real_sha1[20];
1421 enum object_type obj_type;
1422 struct git_istream *st;
1423 git_SHA_CTX c;
1424 char hdr[32];
1425 int hdrlen;
1426
1427 if (map) {
1428 hash_sha1_file(map, size, type, real_sha1);
1429 return hashcmp(sha1, real_sha1) ? -1 : 0;
1430 }
1431
1432 st = open_istream(sha1, &obj_type, &size, NULL);
1433 if (!st)
1434 return -1;
1435
1436 /* Generate the header */
1437 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1438
1439 /* Sha1.. */
1440 git_SHA1_Init(&c);
1441 git_SHA1_Update(&c, hdr, hdrlen);
1442 for (;;) {
1443 char buf[1024 * 16];
1444 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1445
1446 if (readlen < 0) {
1447 close_istream(st);
1448 return -1;
1449 }
1450 if (!readlen)
1451 break;
1452 git_SHA1_Update(&c, buf, readlen);
1453 }
1454 git_SHA1_Final(real_sha1, &c);
1455 close_istream(st);
1456 return hashcmp(sha1, real_sha1) ? -1 : 0;
1457 }
1458
1459 int git_open_noatime(const char *name)
1460 {
1461 static int sha1_file_open_flag = O_NOATIME;
1462
1463 for (;;) {
1464 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1465 if (fd >= 0)
1466 return fd;
1467
1468 /* Might the failure be due to O_NOATIME? */
1469 if (errno != ENOENT && sha1_file_open_flag) {
1470 sha1_file_open_flag = 0;
1471 continue;
1472 }
1473
1474 return -1;
1475 }
1476 }
1477
1478 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1479 {
1480 struct alternate_object_database *alt;
1481
1482 if (!lstat(sha1_file_name(sha1), st))
1483 return 0;
1484
1485 prepare_alt_odb();
1486 errno = ENOENT;
1487 for (alt = alt_odb_list; alt; alt = alt->next) {
1488 fill_sha1_path(alt->name, sha1);
1489 if (!lstat(alt->base, st))
1490 return 0;
1491 }
1492
1493 return -1;
1494 }
1495
1496 static int open_sha1_file(const unsigned char *sha1)
1497 {
1498 int fd;
1499 struct alternate_object_database *alt;
1500 int most_interesting_errno;
1501
1502 fd = git_open_noatime(sha1_file_name(sha1));
1503 if (fd >= 0)
1504 return fd;
1505 most_interesting_errno = errno;
1506
1507 prepare_alt_odb();
1508 for (alt = alt_odb_list; alt; alt = alt->next) {
1509 fill_sha1_path(alt->name, sha1);
1510 fd = git_open_noatime(alt->base);
1511 if (fd >= 0)
1512 return fd;
1513 if (most_interesting_errno == ENOENT)
1514 most_interesting_errno = errno;
1515 }
1516 errno = most_interesting_errno;
1517 return -1;
1518 }
1519
1520 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1521 {
1522 void *map;
1523 int fd;
1524
1525 fd = open_sha1_file(sha1);
1526 map = NULL;
1527 if (fd >= 0) {
1528 struct stat st;
1529
1530 if (!fstat(fd, &st)) {
1531 *size = xsize_t(st.st_size);
1532 if (!*size) {
1533 /* mmap() is forbidden on empty files */
1534 error("object file %s is empty", sha1_file_name(sha1));
1535 return NULL;
1536 }
1537 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1538 }
1539 close(fd);
1540 }
1541 return map;
1542 }
1543
1544 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1545 unsigned long len, enum object_type *type, unsigned long *sizep)
1546 {
1547 unsigned shift;
1548 unsigned long size, c;
1549 unsigned long used = 0;
1550
1551 c = buf[used++];
1552 *type = (c >> 4) & 7;
1553 size = c & 15;
1554 shift = 4;
1555 while (c & 0x80) {
1556 if (len <= used || bitsizeof(long) <= shift) {
1557 error("bad object header");
1558 size = used = 0;
1559 break;
1560 }
1561 c = buf[used++];
1562 size += (c & 0x7f) << shift;
1563 shift += 7;
1564 }
1565 *sizep = size;
1566 return used;
1567 }
1568
1569 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1570 {
1571 /* Get the data stream */
1572 memset(stream, 0, sizeof(*stream));
1573 stream->next_in = map;
1574 stream->avail_in = mapsize;
1575 stream->next_out = buffer;
1576 stream->avail_out = bufsiz;
1577
1578 git_inflate_init(stream);
1579 return git_inflate(stream, 0);
1580 }
1581
1582 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1583 unsigned long mapsize, void *buffer,
1584 unsigned long bufsiz, struct strbuf *header)
1585 {
1586 int status;
1587
1588 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1589
1590 /*
1591 * Check if entire header is unpacked in the first iteration.
1592 */
1593 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1594 return 0;
1595
1596 /*
1597 * buffer[0..bufsiz] was not large enough. Copy the partial
1598 * result out to header, and then append the result of further
1599 * reading the stream.
1600 */
1601 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1602 stream->next_out = buffer;
1603 stream->avail_out = bufsiz;
1604
1605 do {
1606 status = git_inflate(stream, 0);
1607 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1608 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1609 return 0;
1610 stream->next_out = buffer;
1611 stream->avail_out = bufsiz;
1612 } while (status != Z_STREAM_END);
1613 return -1;
1614 }
1615
1616 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1617 {
1618 int bytes = strlen(buffer) + 1;
1619 unsigned char *buf = xmallocz(size);
1620 unsigned long n;
1621 int status = Z_OK;
1622
1623 n = stream->total_out - bytes;
1624 if (n > size)
1625 n = size;
1626 memcpy(buf, (char *) buffer + bytes, n);
1627 bytes = n;
1628 if (bytes <= size) {
1629 /*
1630 * The above condition must be (bytes <= size), not
1631 * (bytes < size). In other words, even though we
1632 * expect no more output and set avail_out to zero,
1633 * the input zlib stream may have bytes that express
1634 * "this concludes the stream", and we *do* want to
1635 * eat that input.
1636 *
1637 * Otherwise we would not be able to test that we
1638 * consumed all the input to reach the expected size;
1639 * we also want to check that zlib tells us that all
1640 * went well with status == Z_STREAM_END at the end.
1641 */
1642 stream->next_out = buf + bytes;
1643 stream->avail_out = size - bytes;
1644 while (status == Z_OK)
1645 status = git_inflate(stream, Z_FINISH);
1646 }
1647 if (status == Z_STREAM_END && !stream->avail_in) {
1648 git_inflate_end(stream);
1649 return buf;
1650 }
1651
1652 if (status < 0)
1653 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1654 else if (stream->avail_in)
1655 error("garbage at end of loose object '%s'",
1656 sha1_to_hex(sha1));
1657 free(buf);
1658 return NULL;
1659 }
1660
1661 /*
1662 * We used to just use "sscanf()", but that's actually way
1663 * too permissive for what we want to check. So do an anal
1664 * object header parse by hand.
1665 */
1666 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1667 unsigned int flags)
1668 {
1669 const char *type_buf = hdr;
1670 unsigned long size;
1671 int type, type_len = 0;
1672
1673 /*
1674 * The type can be of any size but is followed by
1675 * a space.
1676 */
1677 for (;;) {
1678 char c = *hdr++;
1679 if (c == ' ')
1680 break;
1681 type_len++;
1682 }
1683
1684 type = type_from_string_gently(type_buf, type_len, 1);
1685 if (oi->typename)
1686 strbuf_add(oi->typename, type_buf, type_len);
1687 /*
1688 * Set type to 0 if its an unknown object and
1689 * we're obtaining the type using '--allow-unkown-type'
1690 * option.
1691 */
1692 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1693 type = 0;
1694 else if (type < 0)
1695 die("invalid object type");
1696 if (oi->typep)
1697 *oi->typep = type;
1698
1699 /*
1700 * The length must follow immediately, and be in canonical
1701 * decimal format (ie "010" is not valid).
1702 */
1703 size = *hdr++ - '0';
1704 if (size > 9)
1705 return -1;
1706 if (size) {
1707 for (;;) {
1708 unsigned long c = *hdr - '0';
1709 if (c > 9)
1710 break;
1711 hdr++;
1712 size = size * 10 + c;
1713 }
1714 }
1715
1716 if (oi->sizep)
1717 *oi->sizep = size;
1718
1719 /*
1720 * The length must be followed by a zero byte
1721 */
1722 return *hdr ? -1 : type;
1723 }
1724
1725 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1726 {
1727 struct object_info oi;
1728
1729 oi.sizep = sizep;
1730 oi.typename = NULL;
1731 oi.typep = NULL;
1732 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1733 }
1734
1735 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1736 {
1737 int ret;
1738 git_zstream stream;
1739 char hdr[8192];
1740
1741 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1742 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1743 return NULL;
1744
1745 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1746 }
1747
1748 unsigned long get_size_from_delta(struct packed_git *p,
1749 struct pack_window **w_curs,
1750 off_t curpos)
1751 {
1752 const unsigned char *data;
1753 unsigned char delta_head[20], *in;
1754 git_zstream stream;
1755 int st;
1756
1757 memset(&stream, 0, sizeof(stream));
1758 stream.next_out = delta_head;
1759 stream.avail_out = sizeof(delta_head);
1760
1761 git_inflate_init(&stream);
1762 do {
1763 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1764 stream.next_in = in;
1765 st = git_inflate(&stream, Z_FINISH);
1766 curpos += stream.next_in - in;
1767 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1768 stream.total_out < sizeof(delta_head));
1769 git_inflate_end(&stream);
1770 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1771 error("delta data unpack-initial failed");
1772 return 0;
1773 }
1774
1775 /* Examine the initial part of the delta to figure out
1776 * the result size.
1777 */
1778 data = delta_head;
1779
1780 /* ignore base size */
1781 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1782
1783 /* Read the result size */
1784 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1785 }
1786
1787 static off_t get_delta_base(struct packed_git *p,
1788 struct pack_window **w_curs,
1789 off_t *curpos,
1790 enum object_type type,
1791 off_t delta_obj_offset)
1792 {
1793 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1794 off_t base_offset;
1795
1796 /* use_pack() assured us we have [base_info, base_info + 20)
1797 * as a range that we can look at without walking off the
1798 * end of the mapped window. Its actually the hash size
1799 * that is assured. An OFS_DELTA longer than the hash size
1800 * is stupid, as then a REF_DELTA would be smaller to store.
1801 */
1802 if (type == OBJ_OFS_DELTA) {
1803 unsigned used = 0;
1804 unsigned char c = base_info[used++];
1805 base_offset = c & 127;
1806 while (c & 128) {
1807 base_offset += 1;
1808 if (!base_offset || MSB(base_offset, 7))
1809 return 0; /* overflow */
1810 c = base_info[used++];
1811 base_offset = (base_offset << 7) + (c & 127);
1812 }
1813 base_offset = delta_obj_offset - base_offset;
1814 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1815 return 0; /* out of bound */
1816 *curpos += used;
1817 } else if (type == OBJ_REF_DELTA) {
1818 /* The base entry _must_ be in the same pack */
1819 base_offset = find_pack_entry_one(base_info, p);
1820 *curpos += 20;
1821 } else
1822 die("I am totally screwed");
1823 return base_offset;
1824 }
1825
1826 /*
1827 * Like get_delta_base above, but we return the sha1 instead of the pack
1828 * offset. This means it is cheaper for REF deltas (we do not have to do
1829 * the final object lookup), but more expensive for OFS deltas (we
1830 * have to load the revidx to convert the offset back into a sha1).
1831 */
1832 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1833 struct pack_window **w_curs,
1834 off_t curpos,
1835 enum object_type type,
1836 off_t delta_obj_offset)
1837 {
1838 if (type == OBJ_REF_DELTA) {
1839 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1840 return base;
1841 } else if (type == OBJ_OFS_DELTA) {
1842 struct revindex_entry *revidx;
1843 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1844 type, delta_obj_offset);
1845
1846 if (!base_offset)
1847 return NULL;
1848
1849 revidx = find_pack_revindex(p, base_offset);
1850 if (!revidx)
1851 return NULL;
1852
1853 return nth_packed_object_sha1(p, revidx->nr);
1854 } else
1855 return NULL;
1856 }
1857
1858 int unpack_object_header(struct packed_git *p,
1859 struct pack_window **w_curs,
1860 off_t *curpos,
1861 unsigned long *sizep)
1862 {
1863 unsigned char *base;
1864 unsigned long left;
1865 unsigned long used;
1866 enum object_type type;
1867
1868 /* use_pack() assures us we have [base, base + 20) available
1869 * as a range that we can look at. (Its actually the hash
1870 * size that is assured.) With our object header encoding
1871 * the maximum deflated object size is 2^137, which is just
1872 * insane, so we know won't exceed what we have been given.
1873 */
1874 base = use_pack(p, w_curs, *curpos, &left);
1875 used = unpack_object_header_buffer(base, left, &type, sizep);
1876 if (!used) {
1877 type = OBJ_BAD;
1878 } else
1879 *curpos += used;
1880
1881 return type;
1882 }
1883
1884 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1885 {
1886 int type;
1887 struct revindex_entry *revidx;
1888 const unsigned char *sha1;
1889 revidx = find_pack_revindex(p, obj_offset);
1890 if (!revidx)
1891 return OBJ_BAD;
1892 sha1 = nth_packed_object_sha1(p, revidx->nr);
1893 mark_bad_packed_object(p, sha1);
1894 type = sha1_object_info(sha1, NULL);
1895 if (type <= OBJ_NONE)
1896 return OBJ_BAD;
1897 return type;
1898 }
1899
1900 #define POI_STACK_PREALLOC 64
1901
1902 static enum object_type packed_to_object_type(struct packed_git *p,
1903 off_t obj_offset,
1904 enum object_type type,
1905 struct pack_window **w_curs,
1906 off_t curpos)
1907 {
1908 off_t small_poi_stack[POI_STACK_PREALLOC];
1909 off_t *poi_stack = small_poi_stack;
1910 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1911
1912 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1913 off_t base_offset;
1914 unsigned long size;
1915 /* Push the object we're going to leave behind */
1916 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1917 poi_stack_alloc = alloc_nr(poi_stack_nr);
1918 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1919 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1920 } else {
1921 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1922 }
1923 poi_stack[poi_stack_nr++] = obj_offset;
1924 /* If parsing the base offset fails, just unwind */
1925 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1926 if (!base_offset)
1927 goto unwind;
1928 curpos = obj_offset = base_offset;
1929 type = unpack_object_header(p, w_curs, &curpos, &size);
1930 if (type <= OBJ_NONE) {
1931 /* If getting the base itself fails, we first
1932 * retry the base, otherwise unwind */
1933 type = retry_bad_packed_offset(p, base_offset);
1934 if (type > OBJ_NONE)
1935 goto out;
1936 goto unwind;
1937 }
1938 }
1939
1940 switch (type) {
1941 case OBJ_BAD:
1942 case OBJ_COMMIT:
1943 case OBJ_TREE:
1944 case OBJ_BLOB:
1945 case OBJ_TAG:
1946 break;
1947 default:
1948 error("unknown object type %i at offset %"PRIuMAX" in %s",
1949 type, (uintmax_t)obj_offset, p->pack_name);
1950 type = OBJ_BAD;
1951 }
1952
1953 out:
1954 if (poi_stack != small_poi_stack)
1955 free(poi_stack);
1956 return type;
1957
1958 unwind:
1959 while (poi_stack_nr) {
1960 obj_offset = poi_stack[--poi_stack_nr];
1961 type = retry_bad_packed_offset(p, obj_offset);
1962 if (type > OBJ_NONE)
1963 goto out;
1964 }
1965 type = OBJ_BAD;
1966 goto out;
1967 }
1968
1969 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1970 struct object_info *oi)
1971 {
1972 struct pack_window *w_curs = NULL;
1973 unsigned long size;
1974 off_t curpos = obj_offset;
1975 enum object_type type;
1976
1977 /*
1978 * We always get the representation type, but only convert it to
1979 * a "real" type later if the caller is interested.
1980 */
1981 type = unpack_object_header(p, &w_curs, &curpos, &size);
1982
1983 if (oi->sizep) {
1984 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1985 off_t tmp_pos = curpos;
1986 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1987 type, obj_offset);
1988 if (!base_offset) {
1989 type = OBJ_BAD;
1990 goto out;
1991 }
1992 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1993 if (*oi->sizep == 0) {
1994 type = OBJ_BAD;
1995 goto out;
1996 }
1997 } else {
1998 *oi->sizep = size;
1999 }
2000 }
2001
2002 if (oi->disk_sizep) {
2003 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2004 *oi->disk_sizep = revidx[1].offset - obj_offset;
2005 }
2006
2007 if (oi->typep) {
2008 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2009 if (*oi->typep < 0) {
2010 type = OBJ_BAD;
2011 goto out;
2012 }
2013 }
2014
2015 if (oi->delta_base_sha1) {
2016 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2017 const unsigned char *base;
2018
2019 base = get_delta_base_sha1(p, &w_curs, curpos,
2020 type, obj_offset);
2021 if (!base) {
2022 type = OBJ_BAD;
2023 goto out;
2024 }
2025
2026 hashcpy(oi->delta_base_sha1, base);
2027 } else
2028 hashclr(oi->delta_base_sha1);
2029 }
2030
2031 out:
2032 unuse_pack(&w_curs);
2033 return type;
2034 }
2035
2036 static void *unpack_compressed_entry(struct packed_git *p,
2037 struct pack_window **w_curs,
2038 off_t curpos,
2039 unsigned long size)
2040 {
2041 int st;
2042 git_zstream stream;
2043 unsigned char *buffer, *in;
2044
2045 buffer = xmallocz_gently(size);
2046 if (!buffer)
2047 return NULL;
2048 memset(&stream, 0, sizeof(stream));
2049 stream.next_out = buffer;
2050 stream.avail_out = size + 1;
2051
2052 git_inflate_init(&stream);
2053 do {
2054 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2055 stream.next_in = in;
2056 st = git_inflate(&stream, Z_FINISH);
2057 if (!stream.avail_out)
2058 break; /* the payload is larger than it should be */
2059 curpos += stream.next_in - in;
2060 } while (st == Z_OK || st == Z_BUF_ERROR);
2061 git_inflate_end(&stream);
2062 if ((st != Z_STREAM_END) || stream.total_out != size) {
2063 free(buffer);
2064 return NULL;
2065 }
2066
2067 return buffer;
2068 }
2069
2070 #define MAX_DELTA_CACHE (256)
2071
2072 static size_t delta_base_cached;
2073
2074 static struct delta_base_cache_lru_list {
2075 struct delta_base_cache_lru_list *prev;
2076 struct delta_base_cache_lru_list *next;
2077 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2078
2079 static struct delta_base_cache_entry {
2080 struct delta_base_cache_lru_list lru;
2081 void *data;
2082 struct packed_git *p;
2083 off_t base_offset;
2084 unsigned long size;
2085 enum object_type type;
2086 } delta_base_cache[MAX_DELTA_CACHE];
2087
2088 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2089 {
2090 unsigned long hash;
2091
2092 hash = (unsigned long)p + (unsigned long)base_offset;
2093 hash += (hash >> 8) + (hash >> 16);
2094 return hash % MAX_DELTA_CACHE;
2095 }
2096
2097 static struct delta_base_cache_entry *
2098 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2099 {
2100 unsigned long hash = pack_entry_hash(p, base_offset);
2101 return delta_base_cache + hash;
2102 }
2103
2104 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2105 struct packed_git *p, off_t base_offset)
2106 {
2107 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2108 }
2109
2110 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2111 {
2112 struct delta_base_cache_entry *ent;
2113 ent = get_delta_base_cache_entry(p, base_offset);
2114 return eq_delta_base_cache_entry(ent, p, base_offset);
2115 }
2116
2117 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2118 {
2119 ent->data = NULL;
2120 ent->lru.next->prev = ent->lru.prev;
2121 ent->lru.prev->next = ent->lru.next;
2122 delta_base_cached -= ent->size;
2123 }
2124
2125 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2126 unsigned long *base_size, enum object_type *type, int keep_cache)
2127 {
2128 struct delta_base_cache_entry *ent;
2129 void *ret;
2130
2131 ent = get_delta_base_cache_entry(p, base_offset);
2132
2133 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2134 return unpack_entry(p, base_offset, type, base_size);
2135
2136 ret = ent->data;
2137
2138 if (!keep_cache)
2139 clear_delta_base_cache_entry(ent);
2140 else
2141 ret = xmemdupz(ent->data, ent->size);
2142 *type = ent->type;
2143 *base_size = ent->size;
2144 return ret;
2145 }
2146
2147 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2148 {
2149 if (ent->data) {
2150 free(ent->data);
2151 ent->data = NULL;
2152 ent->lru.next->prev = ent->lru.prev;
2153 ent->lru.prev->next = ent->lru.next;
2154 delta_base_cached -= ent->size;
2155 }
2156 }
2157
2158 void clear_delta_base_cache(void)
2159 {
2160 unsigned long p;
2161 for (p = 0; p < MAX_DELTA_CACHE; p++)
2162 release_delta_base_cache(&delta_base_cache[p]);
2163 }
2164
2165 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2166 void *base, unsigned long base_size, enum object_type type)
2167 {
2168 unsigned long hash = pack_entry_hash(p, base_offset);
2169 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2170 struct delta_base_cache_lru_list *lru;
2171
2172 release_delta_base_cache(ent);
2173 delta_base_cached += base_size;
2174
2175 for (lru = delta_base_cache_lru.next;
2176 delta_base_cached > delta_base_cache_limit
2177 && lru != &delta_base_cache_lru;
2178 lru = lru->next) {
2179 struct delta_base_cache_entry *f = (void *)lru;
2180 if (f->type == OBJ_BLOB)
2181 release_delta_base_cache(f);
2182 }
2183 for (lru = delta_base_cache_lru.next;
2184 delta_base_cached > delta_base_cache_limit
2185 && lru != &delta_base_cache_lru;
2186 lru = lru->next) {
2187 struct delta_base_cache_entry *f = (void *)lru;
2188 release_delta_base_cache(f);
2189 }
2190
2191 ent->p = p;
2192 ent->base_offset = base_offset;
2193 ent->type = type;
2194 ent->data = base;
2195 ent->size = base_size;
2196 ent->lru.next = &delta_base_cache_lru;
2197 ent->lru.prev = delta_base_cache_lru.prev;
2198 delta_base_cache_lru.prev->next = &ent->lru;
2199 delta_base_cache_lru.prev = &ent->lru;
2200 }
2201
2202 static void *read_object(const unsigned char *sha1, enum object_type *type,
2203 unsigned long *size);
2204
2205 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2206 {
2207 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2208 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2209 p->pack_name, (uintmax_t)obj_offset);
2210 }
2211
2212 int do_check_packed_object_crc;
2213
2214 #define UNPACK_ENTRY_STACK_PREALLOC 64
2215 struct unpack_entry_stack_ent {
2216 off_t obj_offset;
2217 off_t curpos;
2218 unsigned long size;
2219 };
2220
2221 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2222 enum object_type *final_type, unsigned long *final_size)
2223 {
2224 struct pack_window *w_curs = NULL;
2225 off_t curpos = obj_offset;
2226 void *data = NULL;
2227 unsigned long size;
2228 enum object_type type;
2229 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2230 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2231 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2232 int base_from_cache = 0;
2233
2234 write_pack_access_log(p, obj_offset);
2235
2236 /* PHASE 1: drill down to the innermost base object */
2237 for (;;) {
2238 off_t base_offset;
2239 int i;
2240 struct delta_base_cache_entry *ent;
2241
2242 ent = get_delta_base_cache_entry(p, curpos);
2243 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2244 type = ent->type;
2245 data = ent->data;
2246 size = ent->size;
2247 clear_delta_base_cache_entry(ent);
2248 base_from_cache = 1;
2249 break;
2250 }
2251
2252 if (do_check_packed_object_crc && p->index_version > 1) {
2253 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2254 unsigned long len = revidx[1].offset - obj_offset;
2255 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2256 const unsigned char *sha1 =
2257 nth_packed_object_sha1(p, revidx->nr);
2258 error("bad packed object CRC for %s",
2259 sha1_to_hex(sha1));
2260 mark_bad_packed_object(p, sha1);
2261 unuse_pack(&w_curs);
2262 return NULL;
2263 }
2264 }
2265
2266 type = unpack_object_header(p, &w_curs, &curpos, &size);
2267 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2268 break;
2269
2270 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2271 if (!base_offset) {
2272 error("failed to validate delta base reference "
2273 "at offset %"PRIuMAX" from %s",
2274 (uintmax_t)curpos, p->pack_name);
2275 /* bail to phase 2, in hopes of recovery */
2276 data = NULL;
2277 break;
2278 }
2279
2280 /* push object, proceed to base */
2281 if (delta_stack_nr >= delta_stack_alloc
2282 && delta_stack == small_delta_stack) {
2283 delta_stack_alloc = alloc_nr(delta_stack_nr);
2284 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2285 memcpy(delta_stack, small_delta_stack,
2286 sizeof(*delta_stack)*delta_stack_nr);
2287 } else {
2288 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2289 }
2290 i = delta_stack_nr++;
2291 delta_stack[i].obj_offset = obj_offset;
2292 delta_stack[i].curpos = curpos;
2293 delta_stack[i].size = size;
2294
2295 curpos = obj_offset = base_offset;
2296 }
2297
2298 /* PHASE 2: handle the base */
2299 switch (type) {
2300 case OBJ_OFS_DELTA:
2301 case OBJ_REF_DELTA:
2302 if (data)
2303 die("BUG in unpack_entry: left loop at a valid delta");
2304 break;
2305 case OBJ_COMMIT:
2306 case OBJ_TREE:
2307 case OBJ_BLOB:
2308 case OBJ_TAG:
2309 if (!base_from_cache)
2310 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2311 break;
2312 default:
2313 data = NULL;
2314 error("unknown object type %i at offset %"PRIuMAX" in %s",
2315 type, (uintmax_t)obj_offset, p->pack_name);
2316 }
2317
2318 /* PHASE 3: apply deltas in order */
2319
2320 /* invariants:
2321 * 'data' holds the base data, or NULL if there was corruption
2322 */
2323 while (delta_stack_nr) {
2324 void *delta_data;
2325 void *base = data;
2326 unsigned long delta_size, base_size = size;
2327 int i;
2328
2329 data = NULL;
2330
2331 if (base)
2332 add_delta_base_cache(p, obj_offset, base, base_size, type);
2333
2334 if (!base) {
2335 /*
2336 * We're probably in deep shit, but let's try to fetch
2337 * the required base anyway from another pack or loose.
2338 * This is costly but should happen only in the presence
2339 * of a corrupted pack, and is better than failing outright.
2340 */
2341 struct revindex_entry *revidx;
2342 const unsigned char *base_sha1;
2343 revidx = find_pack_revindex(p, obj_offset);
2344 if (revidx) {
2345 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2346 error("failed to read delta base object %s"
2347 " at offset %"PRIuMAX" from %s",
2348 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2349 p->pack_name);
2350 mark_bad_packed_object(p, base_sha1);
2351 base = read_object(base_sha1, &type, &base_size);
2352 }
2353 }
2354
2355 i = --delta_stack_nr;
2356 obj_offset = delta_stack[i].obj_offset;
2357 curpos = delta_stack[i].curpos;
2358 delta_size = delta_stack[i].size;
2359
2360 if (!base)
2361 continue;
2362
2363 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2364
2365 if (!delta_data) {
2366 error("failed to unpack compressed delta "
2367 "at offset %"PRIuMAX" from %s",
2368 (uintmax_t)curpos, p->pack_name);
2369 data = NULL;
2370 continue;
2371 }
2372
2373 data = patch_delta(base, base_size,
2374 delta_data, delta_size,
2375 &size);
2376
2377 /*
2378 * We could not apply the delta; warn the user, but keep going.
2379 * Our failure will be noticed either in the next iteration of
2380 * the loop, or if this is the final delta, in the caller when
2381 * we return NULL. Those code paths will take care of making
2382 * a more explicit warning and retrying with another copy of
2383 * the object.
2384 */
2385 if (!data)
2386 error("failed to apply delta");
2387
2388 free(delta_data);
2389 }
2390
2391 *final_type = type;
2392 *final_size = size;
2393
2394 unuse_pack(&w_curs);
2395
2396 if (delta_stack != small_delta_stack)
2397 free(delta_stack);
2398
2399 return data;
2400 }
2401
2402 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2403 uint32_t n)
2404 {
2405 const unsigned char *index = p->index_data;
2406 if (!index) {
2407 if (open_pack_index(p))
2408 return NULL;
2409 index = p->index_data;
2410 }
2411 if (n >= p->num_objects)
2412 return NULL;
2413 index += 4 * 256;
2414 if (p->index_version == 1) {
2415 return index + 24 * n + 4;
2416 } else {
2417 index += 8;
2418 return index + 20 * n;
2419 }
2420 }
2421
2422 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2423 {
2424 const unsigned char *index = p->index_data;
2425 index += 4 * 256;
2426 if (p->index_version == 1) {
2427 return ntohl(*((uint32_t *)(index + 24 * n)));
2428 } else {
2429 uint32_t off;
2430 index += 8 + p->num_objects * (20 + 4);
2431 off = ntohl(*((uint32_t *)(index + 4 * n)));
2432 if (!(off & 0x80000000))
2433 return off;
2434 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2435 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2436 ntohl(*((uint32_t *)(index + 4)));
2437 }
2438 }
2439
2440 off_t find_pack_entry_one(const unsigned char *sha1,
2441 struct packed_git *p)
2442 {
2443 const uint32_t *level1_ofs = p->index_data;
2444 const unsigned char *index = p->index_data;
2445 unsigned hi, lo, stride;
2446 static int use_lookup = -1;
2447 static int debug_lookup = -1;
2448
2449 if (debug_lookup < 0)
2450 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2451
2452 if (!index) {
2453 if (open_pack_index(p))
2454 return 0;
2455 level1_ofs = p->index_data;
2456 index = p->index_data;
2457 }
2458 if (p->index_version > 1) {
2459 level1_ofs += 2;
2460 index += 8;
2461 }
2462 index += 4 * 256;
2463 hi = ntohl(level1_ofs[*sha1]);
2464 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2465 if (p->index_version > 1) {
2466 stride = 20;
2467 } else {
2468 stride = 24;
2469 index += 4;
2470 }
2471
2472 if (debug_lookup)
2473 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2474 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2475
2476 if (use_lookup < 0)
2477 use_lookup = !!getenv("GIT_USE_LOOKUP");
2478 if (use_lookup) {
2479 int pos = sha1_entry_pos(index, stride, 0,
2480 lo, hi, p->num_objects, sha1);
2481 if (pos < 0)
2482 return 0;
2483 return nth_packed_object_offset(p, pos);
2484 }
2485
2486 do {
2487 unsigned mi = (lo + hi) / 2;
2488 int cmp = hashcmp(index + mi * stride, sha1);
2489
2490 if (debug_lookup)
2491 printf("lo %u hi %u rg %u mi %u\n",
2492 lo, hi, hi - lo, mi);
2493 if (!cmp)
2494 return nth_packed_object_offset(p, mi);
2495 if (cmp > 0)
2496 hi = mi;
2497 else
2498 lo = mi+1;
2499 } while (lo < hi);
2500 return 0;
2501 }
2502
2503 int is_pack_valid(struct packed_git *p)
2504 {
2505 /* An already open pack is known to be valid. */
2506 if (p->pack_fd != -1)
2507 return 1;
2508
2509 /* If the pack has one window completely covering the
2510 * file size, the pack is known to be valid even if
2511 * the descriptor is not currently open.
2512 */
2513 if (p->windows) {
2514 struct pack_window *w = p->windows;
2515
2516 if (!w->offset && w->len == p->pack_size)
2517 return 1;
2518 }
2519
2520 /* Force the pack to open to prove its valid. */
2521 return !open_packed_git(p);
2522 }
2523
2524 static int fill_pack_entry(const unsigned char *sha1,
2525 struct pack_entry *e,
2526 struct packed_git *p)
2527 {
2528 off_t offset;
2529
2530 if (p->num_bad_objects) {
2531 unsigned i;
2532 for (i = 0; i < p->num_bad_objects; i++)
2533 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2534 return 0;
2535 }
2536
2537 offset = find_pack_entry_one(sha1, p);
2538 if (!offset)
2539 return 0;
2540
2541 /*
2542 * We are about to tell the caller where they can locate the
2543 * requested object. We better make sure the packfile is
2544 * still here and can be accessed before supplying that
2545 * answer, as it may have been deleted since the index was
2546 * loaded!
2547 */
2548 if (!is_pack_valid(p))
2549 return 0;
2550 e->offset = offset;
2551 e->p = p;
2552 hashcpy(e->sha1, sha1);
2553 return 1;
2554 }
2555
2556 /*
2557 * Iff a pack file contains the object named by sha1, return true and
2558 * store its location to e.
2559 */
2560 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2561 {
2562 struct packed_git *p;
2563
2564 prepare_packed_git();
2565 if (!packed_git)
2566 return 0;
2567
2568 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2569 return 1;
2570
2571 for (p = packed_git; p; p = p->next) {
2572 if (p == last_found_pack)
2573 continue; /* we already checked this one */
2574
2575 if (fill_pack_entry(sha1, e, p)) {
2576 last_found_pack = p;
2577 return 1;
2578 }
2579 }
2580 return 0;
2581 }
2582
2583 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2584 struct packed_git *packs)
2585 {
2586 struct packed_git *p;
2587
2588 for (p = packs; p; p = p->next) {
2589 if (find_pack_entry_one(sha1, p))
2590 return p;
2591 }
2592 return NULL;
2593
2594 }
2595
2596 static int sha1_loose_object_info(const unsigned char *sha1,
2597 struct object_info *oi,
2598 int flags)
2599 {
2600 int status = 0;
2601 unsigned long mapsize;
2602 void *map;
2603 git_zstream stream;
2604 char hdr[32];
2605 struct strbuf hdrbuf = STRBUF_INIT;
2606
2607 if (oi->delta_base_sha1)
2608 hashclr(oi->delta_base_sha1);
2609
2610 /*
2611 * If we don't care about type or size, then we don't
2612 * need to look inside the object at all. Note that we
2613 * do not optimize out the stat call, even if the
2614 * caller doesn't care about the disk-size, since our
2615 * return value implicitly indicates whether the
2616 * object even exists.
2617 */
2618 if (!oi->typep && !oi->typename && !oi->sizep) {
2619 struct stat st;
2620 if (stat_sha1_file(sha1, &st) < 0)
2621 return -1;
2622 if (oi->disk_sizep)
2623 *oi->disk_sizep = st.st_size;
2624 return 0;
2625 }
2626
2627 map = map_sha1_file(sha1, &mapsize);
2628 if (!map)
2629 return -1;
2630 if (oi->disk_sizep)
2631 *oi->disk_sizep = mapsize;
2632 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2633 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2634 status = error("unable to unpack %s header with --allow-unknown-type",
2635 sha1_to_hex(sha1));
2636 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2637 status = error("unable to unpack %s header",
2638 sha1_to_hex(sha1));
2639 if (status < 0)
2640 ; /* Do nothing */
2641 else if (hdrbuf.len) {
2642 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2643 status = error("unable to parse %s header with --allow-unknown-type",
2644 sha1_to_hex(sha1));
2645 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2646 status = error("unable to parse %s header", sha1_to_hex(sha1));
2647 git_inflate_end(&stream);
2648 munmap(map, mapsize);
2649 if (status && oi->typep)
2650 *oi->typep = status;
2651 strbuf_release(&hdrbuf);
2652 return 0;
2653 }
2654
2655 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2656 {
2657 struct cached_object *co;
2658 struct pack_entry e;
2659 int rtype;
2660 enum object_type real_type;
2661 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2662
2663 co = find_cached_object(real);
2664 if (co) {
2665 if (oi->typep)
2666 *(oi->typep) = co->type;
2667 if (oi->sizep)
2668 *(oi->sizep) = co->size;
2669 if (oi->disk_sizep)
2670 *(oi->disk_sizep) = 0;
2671 if (oi->delta_base_sha1)
2672 hashclr(oi->delta_base_sha1);
2673 if (oi->typename)
2674 strbuf_addstr(oi->typename, typename(co->type));
2675 oi->whence = OI_CACHED;
2676 return 0;
2677 }
2678
2679 if (!find_pack_entry(real, &e)) {
2680 /* Most likely it's a loose object. */
2681 if (!sha1_loose_object_info(real, oi, flags)) {
2682 oi->whence = OI_LOOSE;
2683 return 0;
2684 }
2685
2686 /* Not a loose object; someone else may have just packed it. */
2687 reprepare_packed_git();
2688 if (!find_pack_entry(real, &e))
2689 return -1;
2690 }
2691
2692 /*
2693 * packed_object_info() does not follow the delta chain to
2694 * find out the real type, unless it is given oi->typep.
2695 */
2696 if (oi->typename && !oi->typep)
2697 oi->typep = &real_type;
2698
2699 rtype = packed_object_info(e.p, e.offset, oi);
2700 if (rtype < 0) {
2701 mark_bad_packed_object(e.p, real);
2702 if (oi->typep == &real_type)
2703 oi->typep = NULL;
2704 return sha1_object_info_extended(real, oi, 0);
2705 } else if (in_delta_base_cache(e.p, e.offset)) {
2706 oi->whence = OI_DBCACHED;
2707 } else {
2708 oi->whence = OI_PACKED;
2709 oi->u.packed.offset = e.offset;
2710 oi->u.packed.pack = e.p;
2711 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2712 rtype == OBJ_OFS_DELTA);
2713 }
2714 if (oi->typename)
2715 strbuf_addstr(oi->typename, typename(*oi->typep));
2716 if (oi->typep == &real_type)
2717 oi->typep = NULL;
2718
2719 return 0;
2720 }
2721
2722 /* returns enum object_type or negative */
2723 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2724 {
2725 enum object_type type;
2726 struct object_info oi = {NULL};
2727
2728 oi.typep = &type;
2729 oi.sizep = sizep;
2730 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2731 return -1;
2732 return type;
2733 }
2734
2735 static void *read_packed_sha1(const unsigned char *sha1,
2736 enum object_type *type, unsigned long *size)
2737 {
2738 struct pack_entry e;
2739 void *data;
2740
2741 if (!find_pack_entry(sha1, &e))
2742 return NULL;
2743 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2744 if (!data) {
2745 /*
2746 * We're probably in deep shit, but let's try to fetch
2747 * the required object anyway from another pack or loose.
2748 * This should happen only in the presence of a corrupted
2749 * pack, and is better than failing outright.
2750 */
2751 error("failed to read object %s at offset %"PRIuMAX" from %s",
2752 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2753 mark_bad_packed_object(e.p, sha1);
2754 data = read_object(sha1, type, size);
2755 }
2756 return data;
2757 }
2758
2759 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2760 unsigned char *sha1)
2761 {
2762 struct cached_object *co;
2763
2764 hash_sha1_file(buf, len, typename(type), sha1);
2765 if (has_sha1_file(sha1) || find_cached_object(sha1))
2766 return 0;
2767 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2768 co = &cached_objects[cached_object_nr++];
2769 co->size = len;
2770 co->type = type;
2771 co->buf = xmalloc(len);
2772 memcpy(co->buf, buf, len);
2773 hashcpy(co->sha1, sha1);
2774 return 0;
2775 }
2776
2777 static void *read_object(const unsigned char *sha1, enum object_type *type,
2778 unsigned long *size)
2779 {
2780 unsigned long mapsize;
2781 void *map, *buf;
2782 struct cached_object *co;
2783
2784 co = find_cached_object(sha1);
2785 if (co) {
2786 *type = co->type;
2787 *size = co->size;
2788 return xmemdupz(co->buf, co->size);
2789 }
2790
2791 buf = read_packed_sha1(sha1, type, size);
2792 if (buf)
2793 return buf;
2794 map = map_sha1_file(sha1, &mapsize);
2795 if (map) {
2796 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2797 munmap(map, mapsize);
2798 return buf;
2799 }
2800 reprepare_packed_git();
2801 return read_packed_sha1(sha1, type, size);
2802 }
2803
2804 /*
2805 * This function dies on corrupt objects; the callers who want to
2806 * deal with them should arrange to call read_object() and give error
2807 * messages themselves.
2808 */
2809 void *read_sha1_file_extended(const unsigned char *sha1,
2810 enum object_type *type,
2811 unsigned long *size,
2812 unsigned flag)
2813 {
2814 void *data;
2815 const struct packed_git *p;
2816 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2817
2818 errno = 0;
2819 data = read_object(repl, type, size);
2820 if (data)
2821 return data;
2822
2823 if (errno && errno != ENOENT)
2824 die_errno("failed to read object %s", sha1_to_hex(sha1));
2825
2826 /* die if we replaced an object with one that does not exist */
2827 if (repl != sha1)
2828 die("replacement %s not found for %s",
2829 sha1_to_hex(repl), sha1_to_hex(sha1));
2830
2831 if (has_loose_object(repl)) {
2832 const char *path = sha1_file_name(sha1);
2833
2834 die("loose object %s (stored in %s) is corrupt",
2835 sha1_to_hex(repl), path);
2836 }
2837
2838 if ((p = has_packed_and_bad(repl)) != NULL)
2839 die("packed object %s (stored in %s) is corrupt",
2840 sha1_to_hex(repl), p->pack_name);
2841
2842 return NULL;
2843 }
2844
2845 void *read_object_with_reference(const unsigned char *sha1,
2846 const char *required_type_name,
2847 unsigned long *size,
2848 unsigned char *actual_sha1_return)
2849 {
2850 enum object_type type, required_type;
2851 void *buffer;
2852 unsigned long isize;
2853 unsigned char actual_sha1[20];
2854
2855 required_type = type_from_string(required_type_name);
2856 hashcpy(actual_sha1, sha1);
2857 while (1) {
2858 int ref_length = -1;
2859 const char *ref_type = NULL;
2860
2861 buffer = read_sha1_file(actual_sha1, &type, &isize);
2862 if (!buffer)
2863 return NULL;
2864 if (type == required_type) {
2865 *size = isize;
2866 if (actual_sha1_return)
2867 hashcpy(actual_sha1_return, actual_sha1);
2868 return buffer;
2869 }
2870 /* Handle references */
2871 else if (type == OBJ_COMMIT)
2872 ref_type = "tree ";
2873 else if (type == OBJ_TAG)
2874 ref_type = "object ";
2875 else {
2876 free(buffer);
2877 return NULL;
2878 }
2879 ref_length = strlen(ref_type);
2880
2881 if (ref_length + 40 > isize ||
2882 memcmp(buffer, ref_type, ref_length) ||
2883 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2884 free(buffer);
2885 return NULL;
2886 }
2887 free(buffer);
2888 /* Now we have the ID of the referred-to object in
2889 * actual_sha1. Check again. */
2890 }
2891 }
2892
2893 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2894 const char *type, unsigned char *sha1,
2895 char *hdr, int *hdrlen)
2896 {
2897 git_SHA_CTX c;
2898
2899 /* Generate the header */
2900 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2901
2902 /* Sha1.. */
2903 git_SHA1_Init(&c);
2904 git_SHA1_Update(&c, hdr, *hdrlen);
2905 git_SHA1_Update(&c, buf, len);
2906 git_SHA1_Final(sha1, &c);
2907 }
2908
2909 /*
2910 * Move the just written object into its final resting place.
2911 */
2912 int finalize_object_file(const char *tmpfile, const char *filename)
2913 {
2914 int ret = 0;
2915
2916 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2917 goto try_rename;
2918 else if (link(tmpfile, filename))
2919 ret = errno;
2920
2921 /*
2922 * Coda hack - coda doesn't like cross-directory links,
2923 * so we fall back to a rename, which will mean that it
2924 * won't be able to check collisions, but that's not a
2925 * big deal.
2926 *
2927 * The same holds for FAT formatted media.
2928 *
2929 * When this succeeds, we just return. We have nothing
2930 * left to unlink.
2931 */
2932 if (ret && ret != EEXIST) {
2933 try_rename:
2934 if (!rename(tmpfile, filename))
2935 goto out;
2936 ret = errno;
2937 }
2938 unlink_or_warn(tmpfile);
2939 if (ret) {
2940 if (ret != EEXIST) {
2941 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2942 }
2943 /* FIXME!!! Collision check here ? */
2944 }
2945
2946 out:
2947 if (adjust_shared_perm(filename))
2948 return error("unable to set permission to '%s'", filename);
2949 return 0;
2950 }
2951
2952 static int write_buffer(int fd, const void *buf, size_t len)
2953 {
2954 if (write_in_full(fd, buf, len) < 0)
2955 return error("file write error (%s)", strerror(errno));
2956 return 0;
2957 }
2958
2959 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2960 unsigned char *sha1)
2961 {
2962 char hdr[32];
2963 int hdrlen;
2964 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2965 return 0;
2966 }
2967
2968 /* Finalize a file on disk, and close it. */
2969 static void close_sha1_file(int fd)
2970 {
2971 if (fsync_object_files)
2972 fsync_or_die(fd, "sha1 file");
2973 if (close(fd) != 0)
2974 die_errno("error when closing sha1 file");
2975 }
2976
2977 /* Size of directory component, including the ending '/' */
2978 static inline int directory_size(const char *filename)
2979 {
2980 const char *s = strrchr(filename, '/');
2981 if (!s)
2982 return 0;
2983 return s - filename + 1;
2984 }
2985
2986 /*
2987 * This creates a temporary file in the same directory as the final
2988 * 'filename'
2989 *
2990 * We want to avoid cross-directory filename renames, because those
2991 * can have problems on various filesystems (FAT, NFS, Coda).
2992 */
2993 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2994 {
2995 int fd, dirlen = directory_size(filename);
2996
2997 if (dirlen + 20 > bufsiz) {
2998 errno = ENAMETOOLONG;
2999 return -1;
3000 }
3001 memcpy(buffer, filename, dirlen);
3002 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
3003 fd = git_mkstemp_mode(buffer, 0444);
3004 if (fd < 0 && dirlen && errno == ENOENT) {
3005 /* Make sure the directory exists */
3006 memcpy(buffer, filename, dirlen);
3007 buffer[dirlen-1] = 0;
3008 if (mkdir(buffer, 0777) && errno != EEXIST)
3009 return -1;
3010 if (adjust_shared_perm(buffer))
3011 return -1;
3012
3013 /* Try again */
3014 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
3015 fd = git_mkstemp_mode(buffer, 0444);
3016 }
3017 return fd;
3018 }
3019
3020 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3021 const void *buf, unsigned long len, time_t mtime)
3022 {
3023 int fd, ret;
3024 unsigned char compressed[4096];
3025 git_zstream stream;
3026 git_SHA_CTX c;
3027 unsigned char parano_sha1[20];
3028 static char tmp_file[PATH_MAX];
3029 const char *filename = sha1_file_name(sha1);
3030
3031 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
3032 if (fd < 0) {
3033 if (errno == EACCES)
3034 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3035 else
3036 return error("unable to create temporary file: %s", strerror(errno));
3037 }
3038
3039 /* Set it up */
3040 git_deflate_init(&stream, zlib_compression_level);
3041 stream.next_out = compressed;
3042 stream.avail_out = sizeof(compressed);
3043 git_SHA1_Init(&c);
3044
3045 /* First header.. */
3046 stream.next_in = (unsigned char *)hdr;
3047 stream.avail_in = hdrlen;
3048 while (git_deflate(&stream, 0) == Z_OK)
3049 ; /* nothing */
3050 git_SHA1_Update(&c, hdr, hdrlen);
3051
3052 /* Then the data itself.. */
3053 stream.next_in = (void *)buf;
3054 stream.avail_in = len;
3055 do {
3056 unsigned char *in0 = stream.next_in;
3057 ret = git_deflate(&stream, Z_FINISH);
3058 git_SHA1_Update(&c, in0, stream.next_in - in0);
3059 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3060 die("unable to write sha1 file");
3061 stream.next_out = compressed;
3062 stream.avail_out = sizeof(compressed);
3063 } while (ret == Z_OK);
3064
3065 if (ret != Z_STREAM_END)
3066 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3067 ret = git_deflate_end_gently(&stream);
3068 if (ret != Z_OK)
3069 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3070 git_SHA1_Final(parano_sha1, &c);
3071 if (hashcmp(sha1, parano_sha1) != 0)
3072 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3073
3074 close_sha1_file(fd);
3075
3076 if (mtime) {
3077 struct utimbuf utb;
3078 utb.actime = mtime;
3079 utb.modtime = mtime;
3080 if (utime(tmp_file, &utb) < 0)
3081 warning("failed utime() on %s: %s",
3082 tmp_file, strerror(errno));
3083 }
3084
3085 return finalize_object_file(tmp_file, filename);
3086 }
3087
3088 static int freshen_loose_object(const unsigned char *sha1)
3089 {
3090 return check_and_freshen(sha1, 1);
3091 }
3092
3093 static int freshen_packed_object(const unsigned char *sha1)
3094 {
3095 struct pack_entry e;
3096 if (!find_pack_entry(sha1, &e))
3097 return 0;
3098 if (e.p->freshened)
3099 return 1;
3100 if (!freshen_file(e.p->pack_name))
3101 return 0;
3102 e.p->freshened = 1;
3103 return 1;
3104 }
3105
3106 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3107 {
3108 char hdr[32];
3109 int hdrlen;
3110
3111 /* Normally if we have it in the pack then we do not bother writing
3112 * it out into .git/objects/??/?{38} file.
3113 */
3114 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3115 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3116 return 0;
3117 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3118 }
3119
3120 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3121 unsigned char *sha1, unsigned flags)
3122 {
3123 char *header;
3124 int hdrlen, status = 0;
3125
3126 /* type string, SP, %lu of the length plus NUL must fit this */
3127 header = xmalloc(strlen(type) + 32);
3128 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3129
3130 if (!(flags & HASH_WRITE_OBJECT))
3131 goto cleanup;
3132 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3133 goto cleanup;
3134 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3135
3136 cleanup:
3137 free(header);
3138 return status;
3139 }
3140
3141 int force_object_loose(const unsigned char *sha1, time_t mtime)
3142 {
3143 void *buf;
3144 unsigned long len;
3145 enum object_type type;
3146 char hdr[32];
3147 int hdrlen;
3148 int ret;
3149
3150 if (has_loose_object(sha1))
3151 return 0;
3152 buf = read_packed_sha1(sha1, &type, &len);
3153 if (!buf)
3154 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3155 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
3156 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3157 free(buf);
3158
3159 return ret;
3160 }
3161
3162 int has_pack_index(const unsigned char *sha1)
3163 {
3164 struct stat st;
3165 if (stat(sha1_pack_index_name(sha1), &st))
3166 return 0;
3167 return 1;
3168 }
3169
3170 int has_sha1_pack(const unsigned char *sha1)
3171 {
3172 struct pack_entry e;
3173 return find_pack_entry(sha1, &e);
3174 }
3175
3176 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3177 {
3178 struct pack_entry e;
3179
3180 if (find_pack_entry(sha1, &e))
3181 return 1;
3182 if (has_loose_object(sha1))
3183 return 1;
3184 if (flags & HAS_SHA1_QUICK)
3185 return 0;
3186 reprepare_packed_git();
3187 return find_pack_entry(sha1, &e);
3188 }
3189
3190 static void check_tree(const void *buf, size_t size)
3191 {
3192 struct tree_desc desc;
3193 struct name_entry entry;
3194
3195 init_tree_desc(&desc, buf, size);
3196 while (tree_entry(&desc, &entry))
3197 /* do nothing
3198 * tree_entry() will die() on malformed entries */
3199 ;
3200 }
3201
3202 static void check_commit(const void *buf, size_t size)
3203 {
3204 struct commit c;
3205 memset(&c, 0, sizeof(c));
3206 if (parse_commit_buffer(&c, buf, size))
3207 die("corrupt commit");
3208 }
3209
3210 static void check_tag(const void *buf, size_t size)
3211 {
3212 struct tag t;
3213 memset(&t, 0, sizeof(t));
3214 if (parse_tag_buffer(&t, buf, size))
3215 die("corrupt tag");
3216 }
3217
3218 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3219 enum object_type type,
3220 const char *path, unsigned flags)
3221 {
3222 int ret, re_allocated = 0;
3223 int write_object = flags & HASH_WRITE_OBJECT;
3224
3225 if (!type)
3226 type = OBJ_BLOB;
3227
3228 /*
3229 * Convert blobs to git internal format
3230 */
3231 if ((type == OBJ_BLOB) && path) {
3232 struct strbuf nbuf = STRBUF_INIT;
3233 if (convert_to_git(path, buf, size, &nbuf,
3234 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3235 buf = strbuf_detach(&nbuf, &size);
3236 re_allocated = 1;
3237 }
3238 }
3239 if (flags & HASH_FORMAT_CHECK) {
3240 if (type == OBJ_TREE)
3241 check_tree(buf, size);
3242 if (type == OBJ_COMMIT)
3243 check_commit(buf, size);
3244 if (type == OBJ_TAG)
3245 check_tag(buf, size);
3246 }
3247
3248 if (write_object)
3249 ret = write_sha1_file(buf, size, typename(type), sha1);
3250 else
3251 ret = hash_sha1_file(buf, size, typename(type), sha1);
3252 if (re_allocated)
3253 free(buf);
3254 return ret;
3255 }
3256
3257 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3258 const char *path, unsigned flags)
3259 {
3260 int ret;
3261 const int write_object = flags & HASH_WRITE_OBJECT;
3262 struct strbuf sbuf = STRBUF_INIT;
3263
3264 assert(path);
3265 assert(would_convert_to_git_filter_fd(path));
3266
3267 convert_to_git_filter_fd(path, fd, &sbuf,
3268 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3269
3270 if (write_object)
3271 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3272 sha1);
3273 else
3274 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3275 sha1);
3276 strbuf_release(&sbuf);
3277 return ret;
3278 }
3279
3280 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3281 const char *path, unsigned flags)
3282 {
3283 struct strbuf sbuf = STRBUF_INIT;
3284 int ret;
3285
3286 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3287 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3288 else
3289 ret = -1;
3290 strbuf_release(&sbuf);
3291 return ret;
3292 }
3293
3294 #define SMALL_FILE_SIZE (32*1024)
3295
3296 static int index_core(unsigned char *sha1, int fd, size_t size,
3297 enum object_type type, const char *path,
3298 unsigned flags)
3299 {
3300 int ret;
3301
3302 if (!size) {
3303 ret = index_mem(sha1, "", size, type, path, flags);
3304 } else if (size <= SMALL_FILE_SIZE) {
3305 char *buf = xmalloc(size);
3306 if (size == read_in_full(fd, buf, size))
3307 ret = index_mem(sha1, buf, size, type, path, flags);
3308 else
3309 ret = error("short read %s", strerror(errno));
3310 free(buf);
3311 } else {
3312 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3313 ret = index_mem(sha1, buf, size, type, path, flags);
3314 munmap(buf, size);
3315 }
3316 return ret;
3317 }
3318
3319 /*
3320 * This creates one packfile per large blob unless bulk-checkin
3321 * machinery is "plugged".
3322 *
3323 * This also bypasses the usual "convert-to-git" dance, and that is on
3324 * purpose. We could write a streaming version of the converting
3325 * functions and insert that before feeding the data to fast-import
3326 * (or equivalent in-core API described above). However, that is
3327 * somewhat complicated, as we do not know the size of the filter
3328 * result, which we need to know beforehand when writing a git object.
3329 * Since the primary motivation for trying to stream from the working
3330 * tree file and to avoid mmaping it in core is to deal with large
3331 * binary blobs, they generally do not want to get any conversion, and
3332 * callers should avoid this code path when filters are requested.
3333 */
3334 static int index_stream(unsigned char *sha1, int fd, size_t size,
3335 enum object_type type, const char *path,
3336 unsigned flags)
3337 {
3338 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3339 }
3340
3341 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3342 enum object_type type, const char *path, unsigned flags)
3343 {
3344 int ret;
3345
3346 /*
3347 * Call xsize_t() only when needed to avoid potentially unnecessary
3348 * die() for large files.
3349 */
3350 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3351 ret = index_stream_convert_blob(sha1, fd, path, flags);
3352 else if (!S_ISREG(st->st_mode))
3353 ret = index_pipe(sha1, fd, type, path, flags);
3354 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3355 (path && would_convert_to_git(path)))
3356 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3357 flags);
3358 else
3359 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3360 flags);
3361 close(fd);
3362 return ret;
3363 }
3364
3365 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3366 {
3367 int fd;
3368 struct strbuf sb = STRBUF_INIT;
3369
3370 switch (st->st_mode & S_IFMT) {
3371 case S_IFREG:
3372 fd = open(path, O_RDONLY);
3373 if (fd < 0)
3374 return error("open(\"%s\"): %s", path,
3375 strerror(errno));
3376 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3377 return error("%s: failed to insert into database",
3378 path);
3379 break;
3380 case S_IFLNK:
3381 if (strbuf_readlink(&sb, path, st->st_size)) {
3382 char *errstr = strerror(errno);
3383 return error("readlink(\"%s\"): %s", path,
3384 errstr);
3385 }
3386 if (!(flags & HASH_WRITE_OBJECT))
3387 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3388 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3389 return error("%s: failed to insert into database",
3390 path);
3391 strbuf_release(&sb);
3392 break;
3393 case S_IFDIR:
3394 return resolve_gitlink_ref(path, "HEAD", sha1);
3395 default:
3396 return error("%s: unsupported file type", path);
3397 }
3398 return 0;
3399 }
3400
3401 int read_pack_header(int fd, struct pack_header *header)
3402 {
3403 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3404 /* "eof before pack header was fully read" */
3405 return PH_ERROR_EOF;
3406
3407 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3408 /* "protocol error (pack signature mismatch detected)" */
3409 return PH_ERROR_PACK_SIGNATURE;
3410 if (!pack_version_ok(header->hdr_version))
3411 /* "protocol error (pack version unsupported)" */
3412 return PH_ERROR_PROTOCOL;
3413 return 0;
3414 }
3415
3416 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3417 {
3418 enum object_type type = sha1_object_info(sha1, NULL);
3419 if (type < 0)
3420 die("%s is not a valid object", sha1_to_hex(sha1));
3421 if (type != expect)
3422 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3423 typename(expect));
3424 }
3425
3426 static int for_each_file_in_obj_subdir(int subdir_nr,
3427 struct strbuf *path,
3428 each_loose_object_fn obj_cb,
3429 each_loose_cruft_fn cruft_cb,
3430 each_loose_subdir_fn subdir_cb,
3431 void *data)
3432 {
3433 size_t baselen = path->len;
3434 DIR *dir = opendir(path->buf);
3435 struct dirent *de;
3436 int r = 0;
3437
3438 if (!dir) {
3439 if (errno == ENOENT)
3440 return 0;
3441 return error("unable to open %s: %s", path->buf, strerror(errno));
3442 }
3443
3444 while ((de = readdir(dir))) {
3445 if (is_dot_or_dotdot(de->d_name))
3446 continue;
3447
3448 strbuf_setlen(path, baselen);
3449 strbuf_addf(path, "/%s", de->d_name);
3450
3451 if (strlen(de->d_name) == 38) {
3452 char hex[41];
3453 unsigned char sha1[20];
3454
3455 snprintf(hex, sizeof(hex), "%02x%s",
3456 subdir_nr, de->d_name);
3457 if (!get_sha1_hex(hex, sha1)) {
3458 if (obj_cb) {
3459 r = obj_cb(sha1, path->buf, data);
3460 if (r)
3461 break;
3462 }
3463 continue;
3464 }
3465 }
3466
3467 if (cruft_cb) {
3468 r = cruft_cb(de->d_name, path->buf, data);
3469 if (r)
3470 break;
3471 }
3472 }
3473 strbuf_setlen(path, baselen);
3474
3475 if (!r && subdir_cb)
3476 r = subdir_cb(subdir_nr, path->buf, data);
3477
3478 closedir(dir);
3479 return r;
3480 }
3481
3482 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3483 each_loose_object_fn obj_cb,
3484 each_loose_cruft_fn cruft_cb,
3485 each_loose_subdir_fn subdir_cb,
3486 void *data)
3487 {
3488 size_t baselen = path->len;
3489 int r = 0;
3490 int i;
3491
3492 for (i = 0; i < 256; i++) {
3493 strbuf_addf(path, "/%02x", i);
3494 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3495 subdir_cb, data);
3496 strbuf_setlen(path, baselen);
3497 if (r)
3498 break;
3499 }
3500
3501 return r;
3502 }
3503
3504 int for_each_loose_file_in_objdir(const char *path,
3505 each_loose_object_fn obj_cb,
3506 each_loose_cruft_fn cruft_cb,
3507 each_loose_subdir_fn subdir_cb,
3508 void *data)
3509 {
3510 struct strbuf buf = STRBUF_INIT;
3511 int r;
3512
3513 strbuf_addstr(&buf, path);
3514 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3515 subdir_cb, data);
3516 strbuf_release(&buf);
3517
3518 return r;
3519 }
3520
3521 struct loose_alt_odb_data {
3522 each_loose_object_fn *cb;
3523 void *data;
3524 };
3525
3526 static int loose_from_alt_odb(struct alternate_object_database *alt,
3527 void *vdata)
3528 {
3529 struct loose_alt_odb_data *data = vdata;
3530 struct strbuf buf = STRBUF_INIT;
3531 int r;
3532
3533 /* copy base not including trailing '/' */
3534 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3535 r = for_each_loose_file_in_objdir_buf(&buf,
3536 data->cb, NULL, NULL,
3537 data->data);
3538 strbuf_release(&buf);
3539 return r;
3540 }
3541
3542 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3543 {
3544 struct loose_alt_odb_data alt;
3545 int r;
3546
3547 r = for_each_loose_file_in_objdir(get_object_directory(),
3548 cb, NULL, NULL, data);
3549 if (r)
3550 return r;
3551
3552 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3553 return 0;
3554
3555 alt.cb = cb;
3556 alt.data = data;
3557 return foreach_alt_odb(loose_from_alt_odb, &alt);
3558 }
3559
3560 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3561 {
3562 uint32_t i;
3563 int r = 0;
3564
3565 for (i = 0; i < p->num_objects; i++) {
3566 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3567
3568 if (!sha1)
3569 return error("unable to get sha1 of object %u in %s",
3570 i, p->pack_name);
3571
3572 r = cb(sha1, p, i, data);
3573 if (r)
3574 break;
3575 }
3576 return r;
3577 }
3578
3579 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3580 {
3581 struct packed_git *p;
3582 int r = 0;
3583 int pack_errors = 0;
3584
3585 prepare_packed_git();
3586 for (p = packed_git; p; p = p->next) {
3587 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3588 continue;
3589 if (open_pack_index(p)) {
3590 pack_errors = 1;
3591 continue;
3592 }
3593 r = for_each_object_in_pack(p, cb, data);
3594 if (r)
3595 break;
3596 }
3597 return r ? r : pack_errors;
3598 }